neural areas of activation during clinical reasoning and
TRANSCRIPT
University of Calgary
PRISM: University of Calgary's Digital Repository
Graduate Studies The Vault: Electronic Theses and Dissertations
2015-07-20
Neural Areas of Activation During Clinical Reasoning
and Decision Making
Hruska, Pamela
Hruska, P. (2015). Neural Areas of Activation During Clinical Reasoning and Decision Making
(Unpublished master's thesis). University of Calgary, Calgary, AB. doi:10.11575/PRISM/27078
http://hdl.handle.net/11023/2358
master thesis
University of Calgary graduate students retain copyright ownership and moral rights for their
thesis. You may use this material in any way that is permitted by the Copyright Act or through
licensing that has been assigned to the document. For uses that are not allowable under
copyright legislation or licensing, you are required to seek permission.
Downloaded from PRISM: https://prism.ucalgary.ca
UNIVERSITY OF CALGARY
Neural Areas of Activation During Clinical Reasoning and Decision Making
by
Pamela Hruska
A THESIS
SUBMITTED TO THE FACULTY OF GRADUATE STUDIES
IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE
DEGREE OF MASTER OF SCIENCE
GRADUATE PROGRAM IN MEDICAL SCIENCE
CALGARY, ALBERTA
JULY, 2015
© Pamela Hruska 2015
ii
Preface
Background: Neural areas of activation involved in clinical reasoning and decision making were
assessed using functional magnetic resonance imaging (fMRI) in novice and expert clinicians as
they reasoned through and assigned clinical diagnoses to sixteen different clinical cases (eight
easy, and eight hard).
Results: During the clinical reasoning phase, novices had increased activation in the left anterior
temporal cortex during easy and hard clinical cases, and the prefrontal cortex during hard clinical
cases. There were no significant differences in brain activity between groups during clinical
decision making for the easy cases. During clinical diagnoses on hard cases, novices had
increased left anterior temporal cortex and left ventrolateral prefrontal cortex activation, whereas
experts had increased activations in the right parietal cortex and right dorsolateral and
ventrolateral prefrontal cortex.
Conclusion: Two modifiers of neural activation during clinical reasoning and clinical diagnoses
include clinician level of expertise and task difficulty. Novice clinicians rely more heavily on
semantic memory, when reasoning and as well demand more working memory (WM) when
reasoning through cases. While both novices and experts demand use of the pre frontal cortex
(PFC) during decision making, differences in hemispheric activations could suggest WM and
supporting areas of the PFC evolve from use of semantic, factual knowledge that is rule-based
guided by basic causal explanations in novices, to processes dedicating more attention to
evaluative assessment in experts where comparisons between exemplars with more internal
experiences are used.
iii
Contribution of Authors
Dr. Kent Hecker – Academic supervisor, co-author on all written manuscripts, primary
investigator, project conceptualization, content expertise in medical education, overall project
quality assessment, review and editing of thesis and all manuscripts.
Dr. Olav Krigolson – Co-author on all written manuscripts, project conceptualization, content
expertise in neuroimaging techniques and neuroscience, MRI protocol design, training and
supervision of fMRI methods and data analysis, review and editing of all manuscripts.
Dr. Sylvain Coderre – Co-author on all written manuscripts, provision of clinical cases used in
this research, content expertise in cognition in medical education, medical expertise in
gastroenterology, medical expertise in clinical reasoning and decision making, recruitment of
expert participants, review of thesis proposal and all manuscripts.
Dr. Kevin McLaughlin – Co-author on all written manuscripts, provision of clinical cases used in
this research, content expertise in cognition in medical education, medical expertise in clinical
reasoning and decision making, reviewed and aided in discussion and conclusions, review of all
manuscripts.
Dr. Tanya Beran - Co-author on all written manuscripts, content expertise in medical education,
review and editing of manuscripts.
Dr. Christopher Doig - Co-author on all written manuscripts, medical expertise in clinical
reasoning and decision making, review and editing of manuscripts.
Dr. Bruce Wright - Co-author on all written manuscripts, project funding, recruitment of novice
participants.
Filomeno Cortese – Co-author on all written manuscripts, MRI protocol implementation, MRI
training, scanning participants, data acquisition, data interpretation, and review and editing of all
manuscripts.
iv
Acknowledgements
This thesis was made possible through the support of an amazing network people in my
life. No matter how big or small these contributions may have seemed to you, they were
essential in this accomplishment, and I am very thankful to all of you.
My supervisor, Dr. Kent Hecker, helped me shatter the confines of the world as I knew it
through this work, and encouraged me to step outside my comfort zone to learn in ways I never
anticipated. Kent, the enthusiasm, curiosity, patience, encouragement and support you provided
over the years were anchors for me being successful and kept me going. Thank you for your
belief in my ability throughout.
Dr. Olav Krigolson, without your expertise and exceptional patience with teaching me
how to analyze fMRI data and making sure I was on the right track, the results of this work
would pale in comparison. Thank you so much for all of the time and energy you put in to
supporting me throughout this, it is truly appreciated.
I would like to thank my committee members Dr. Tanya Beran, Dr. Sylvain Coderre and
Dr. Chip Doig for supporting this work in so many ways. Your efforts were essential for helping
to refine this thesis. I would also like to thank the contributors to the manuscripts in this work,
Dr. Kevin McLaughlin, Filomeno Cortese and Dr. Bruce Wright.
Chip Doig, without your encouragement and support, it is possible I would never have
crossed over from the hospital to pursue this academic journey. Thank you for your sage advice,
and for encouraging me on this path.
The support from Dr. Brad Goodyear and the Seaman Family MR Centre not only made
this work possible, but enjoyable. Filomeno Cortese, Dan Pittman, and Francis Raymond, thank
you for your hours of support.
v
I would like to acknowledge all of my clinical colleagues. Joan Harris and Chris
Coltman, thank you for lending your brains to test out our protocols. The critical care clinical
nurse educators, managers, and cardiac CNS I work with were a continued source of
encouragement. Caroline Hatcher, you were quite simply a rock for me throughout this, and I
could not have done this with out you.
Lastly, thank you to all of the participants involved in this work. I am forever thankful
you volunteered to make this possible.
vi
Dedication
There is no way I could have completed this journey without my friends and family. You
are a source of balance in my life. This work is dedicated to your endless love and
encouragement.
vii
Table of Contents
Preface ................................................................................................................................. ii Dedication .......................................................................................................................... vi Table of Contents .............................................................................................................. vii List of Tables .......................................................................................................................x List of Figures and Illustrations ......................................................................................... xi List of Abbreviations ........................................................................................................ xii
CHAPTER ONE: INTRODUCTION ................................................................................13 1.1 Overview ..................................................................................................................13 1.2 Study Purpose ..........................................................................................................15 1.3 Study Objective ........................................................................................................15 1.4 Hypotheses and Predictions .....................................................................................15
1.4.1 Hypotheses related to neural areas of activation in clinical reasoning: ...........16 1.4.2 Predictions related to neural areas of activation in clinical decision making: 16
1.5 Thesis Outline ..........................................................................................................16
CHAPTER TWO: BACKGROUND AND LITERATURE REVIEW .............................18 2.1 Memory Structure and Organization .......................................................................18 2.2 Overview and Definitions of Clinical Reasoning and Clinical Decision Making ...21 2.3 Theories and Methods in Studying Clinical Reasoning and Decision Making .......22 2.4 Knowledge Structure and Expertise ........................................................................23
2.4.1 Novice vs. Expert Clinical Reasoning .............................................................25 2.5 Theoretical Explanations of Clinical Reasoning and Decision-Making Strategies .25
2.5.1 Hypothetico-Deductive (Backwards) Reasoning ............................................25 2.5.2 Inductive and Scheme Inductive (Forward) Reasoning ..................................26 2.5.3 Pattern Recognition and Recognition-Primed Decision Making ....................26 2.5.4 Dual Processing Theories ................................................................................27
2.6 Neuroimaging Studies Conducted within Medical Education .................................28
CHAPTER THREE: BASICS OF FUNCTIONAL MAGNETIC RESONANCE IMAGING 3.1 Basics of MRI and fMRI .........................................................................................32 3.2 Tasks and Experimental Design ..............................................................................33 3.3 Sample Size ..............................................................................................................35 3.4 Pre Processing ..........................................................................................................36
3.4.1 fMRI Data Analysis .........................................................................................37 3.4.2 Statistical Inference .........................................................................................39
3.5 Hardware and Software ...........................................................................................40 3.6 Participant Considerations .......................................................................................41
3.6.1 MRI Safety ......................................................................................................41 3.6.2 Exclusion Criteria ............................................................................................41 3.6.3 Practical Considerations ..................................................................................41
CHAPTER FOUR: MANUSCRIPT # 1 ............................................................................43 4.0 Making up your mind: neural areas of activation during clinical reasoning ...........43 4.1 Abstract ....................................................................................................................43
viii
4.2 Introduction ..............................................................................................................44 4.2.1 Memory and Reasoning ...................................................................................45 4.2.2 fMRI and Medical Education ..........................................................................46
4.3 Purpose .....................................................................................................................47 4.4 Methods ...................................................................................................................47
4.4.1 Participants ......................................................................................................47 4.4.2 Stimuli and Procedures ....................................................................................48 4.4.3 Clinical Cases ..................................................................................................48
4.5 Data Acquisition and Analysis ................................................................................49 4.5.1 Functional and Structural Data Acquisition ....................................................49 4.5.2 fMRI Data Processing and Analysis ................................................................50
4.6 Results ......................................................................................................................51 4.6.1 fMRI ................................................................................................................51
4.7 Discussion ................................................................................................................52 4.7.1 Other Areas of Activation ...............................................................................54
4.8 Limitations ...............................................................................................................55 4.9 Conclusion ...............................................................................................................55
CHAPTER FIVE: MANUSCRIPT # 2 .............................................................................61 5.0 Mind Made Up: Neural areas of activation during clinical decision making ..........61 5.1 Abstract ....................................................................................................................61 5.2 Introduction ..............................................................................................................63
5.2.1 Clinical Decision Making: Stages and Categories ..........................................63 5.2.2 fMRI of Clinical Decision Making ..................................................................64 5.2.3 Neural Areas of Interest ...................................................................................66
5.3 Methods ...................................................................................................................67 5.3.1 Participants ......................................................................................................67 5.3.2 Stimuli and Procedures ....................................................................................67 5.3.3 Clinical Cases ..................................................................................................68
5.4 Data Acquisition and Analysis ................................................................................68 5.4.1 Behavioural Data and Analysis .......................................................................68 5.4.2 Functional and Structural Data Acquisition ....................................................69 5.4.3 fMRI Data Processing and Analysis ................................................................69
5.5 Results ......................................................................................................................71 5.5.1 Behavioral Results ...........................................................................................71 5.5.2 fMRI Results ...................................................................................................71
5.6 Discussion ................................................................................................................72 5.7 Limitations ...............................................................................................................76 5.8 Conclusion ...............................................................................................................77
CHAPTER SIX: SUPPLEMENTAL RESULTS ..............................................................82 6.1 Total Study Recruitment ..........................................................................................82 6.2 Pre Screening Questionnaire Data ...........................................................................82
6.2.1 Education Achievement Levels .......................................................................82 6.2.2 Formal Teaching ..............................................................................................82 6.2.3 Years of Clinical Experience ...........................................................................83 6.2.4 Hours of Sleep .................................................................................................83
ix
6.3 Post Screening Questionnaire Data ..........................................................................83 6.4 Confidence in Decisions ..........................................................................................84
6.4.1 Overall Confidence in Decisions .....................................................................85 6.4.2 Confidence on Easy and Hard Questions ........................................................85
CHAPTER SEVEN: OVERALL DISCUSSION AND CONCLUSIONS .......................86 7.1 Key Findings ............................................................................................................86
7.1.1 Behavioural Findings: .....................................................................................86 7.1.2 Overall Neural Findings: .................................................................................86 7.1.3 Neural Areas of Activation and use of Memory in Clinical Reasoning ..........87 7.1.4 Neural Areas of Activation and use of Memory in Clinical Decision Making87
7.2 Discussion ................................................................................................................87 7.3 Strengths ..................................................................................................................90 7.4 Limitations ...............................................................................................................91 7.5 Future Studies ..........................................................................................................92 7.6 Conclusions ..............................................................................................................93
REFERNCES .....................................................................................................................94
APPENDIX A: MRI SCREENING FORM ...................................................................104
APPENDIX B: PRE SCREENING QUESTIONNAIRE ...............................................105
APPENDIX C: POST SCREENING QUESTIONNAIRE ............................................106
x
List of Tables
Table 2.1 Summary of Memory Components and Neural Correlates .......................................... 20
Table 3.1 Pre Processing fMRI data ............................................................................................ 36
Table 3.2 First Level Parameters of Interest and Onset Times .................................................... 38
Table 3.3 Contrasts of Interest for First Level Participant Analyses ........................................... 38
Table 3.4 Contrasts of Interest for Second Level Group Analyses .............................................. 38
Table 3.5 Contrasts of Interest for Final Level Analyses ............................................................ 39
Table 3.6 Required Hardware and Software for fMRI Research ................................................. 40
Table 4.1 Reading: Common Clusters for Novices Reading Easy and Hard Clinical Cases ...... 58
Table 4.2 Reading: Common Clusters for Experts Reading Easy and Hard Clinical Cases ....... 59
Table 4.3 Reading: Novice > Expert Activations ........................................................................ 60
Table 4.4 Reading: Areas of Activation in Regions of Interest: Novice > Expert; Hard > Easy ....................................................................................................................................... 60
Table 5.1 MCQ: Common Clusters for Novices Diagnosing Easy and Hard Clinical Cases .... 78
Table 5.2 MCQ: Common Clusters for Experts Diagnosing Easy and Hard MCQ .................... 79
Table 5.3 MCQ: Hard Questions: Novice > Expert ..................................................................... 79
Table 5.4 MCQ: Hard Questions: Expert > Novice ..................................................................... 80
Table 6.1 Confidence Answer Options and Associated Ratings and Coding .............................. 84
xi
List of Figures and Illustrations
Figure 2.1 Depiction of Memory Structure Supporting Information Retrieval ............................ 21
Figure 3.1 Timing Diagram and Case Presentation of One Functional Run ............................... 34
Figure 4.1 Combined Neural Areas of Activation in Clinical Reasoning ................................... 56
Figure 4.2 Region of Interest Images in Clinical Reasoning ....................................................... 57
Figure 5.1 Combined Neural Areas of Activation in Clinical Decision Making ......................... 80
Figure 5.2 Novice Expert Differences in Clinical Decision Making on Hard Tasks .................. 81
xii
List of Abbreviations
Abbreviation Definition 3D Three Dimensional ANOVA Analysis of Variance BOLD Blood Oxygen Light Dependent CHREB Calgary Health Ethics Board DLPFC Dorsolateral Prefrontal Cortex DPT Dual Process Theory EEG Electroencephalogram FA Flip Angle fMRI Functional Magnetic Resonance Imaging FOV Field of View GLM General Linear Model GRE-EPI Gradient Recalled Echo, Echo Planar Imaging LTM Long Term Memory MCQ Multiple Choice Question MNI Montreal Neurological Institute MRI Magnetic Resonance Imaging MTL Medial Temporal Lobe NMR Nuclear Magnetic Resonance PFC Prefrontal Cortex PL Phonological Loop RF Radiofrequency ROI Region of Interest TE Echo Time TR Repetition Time VLPFC Ventrolateral Pre Frontal Cortex VSSP Visuospatial Sketch Pad WM Working Memory
13
Chapter One: INTRODUCTION
1.1 Overview
Medical education research has continually sought to understand the neurocognitive
processes underlying clinical reasoning and decision making. The importance of this work was
identified as far back as Elstein 1978, where he states; “early researchers in medical education
remain correct in that we will not be able to dependably teach or evaluate one of the most
important characteristics of being a physician without an understanding of the underlying
processes,”1. To date, indirect methods have predominantly been used to study how physicians
become expert diagnosticians over the course of medical training. These approaches have
included concept sorting2; where participant written answers are analyzed by researchers to draw
inferences about what must have been known by the way they responded, or by think aloud
protocol3-4; a technique requiring participants to concurrently or retrospectively ‘talk through’ or
explain the mental process they used to make a decision. Information gained from these
approaches has helped shape clinical reasoning/decision-making theories, and though it has
afforded great insight, continued use of these research methods have been cautioned against.
Namely, results from these experiments may be a confabulation of how participants explain their
reasoning and final decisions versus a true reflection of the neurocognitive process involved5,6,7.
Neuroimaging studies, using technology such as functional magnetic resonance imaging
(fMRI), could provide a more direct method for assessing how clinicians’ clinically reason and
make decisions8, which in turn can help inform standing theoretical explanations. There are a
few preliminary neuroimaging studies conducted within medical education at present, and
though this body of literature is small, there is great diversity in the ways clinical reasoning and
decision making have been explored. Some have studied tasks which are visuospatial in nature,
14
such as in diagnostic imaging tasks9,10, while others have explored neural areas of activation in
decision making in altered mental states11, in associative learning tasks12, and in relation to
analytical and non-analytical reasoning3,13. Given that retrieval of information from long-term
memory (LTM), made possible by engaging working memory (WM), is what underlies one’s
ability to reason in general14-16, an alternate approach for examining how the brain works when
engaged in clinical reasoning and decision-making tasks is to do so from the framework of
memory structure15,14. Expanded upon further in upcoming chapters of this thesis, WM can be
thought of as a supporting memory structure that temporarily holds relevant information in a
heightened state of availability so that key information required for reasoning/decision making is
readily accessible17. The relevant information is retrieved from LTM, which is the form of
memory structure which holds a vast store of knowledge and a record of prior events18.
Neuroimaging research findings from other disciplines have identified the medial
temporal lobe (MTL)19 and prefrontal cortex (PFC)20 as neural areas associated with WM during
reasoning tasks. The MTL receives multiple sensory inputs and is required for usable long-term
declarative memory21,22, for basic language perception and integration and interpretation of
words and sentences23. The PFC is linked to WM which is responsible for retrieving information
from LTM to be used in language comprehension, planning, assessment, execution and outcome
processing24-26,27,28.
This pilot work is among the first to explore the neural substructures supporting the
cognitive processes of clinical reasoning and decision making within medical education from the
framework of memory structure using fMRI. Given effective and safe patient care are dependent
on both clinical reasoning and decision-making processes being sound, understanding the neural
basis of these cognitive processes is of importance29-30. Through a combined medical education
15
and neuroscientific approach to understanding clinical reasoning and decision making from a
memory structure framework, and by studying different levels of clinical expertise (novice and
expert) and task difficulty (easy and hard), this work will provide increased awareness of what
neural areas support these cognitive processes and aims to expose if and when differential neural
activity occurs1,8,31.
1.2 Study Purpose
The purpose of this pilot fMRI study was to determine the neural areas of activation in
clinical reasoning and clinical decision making, and to identify if clinician level of expertise or
manipulations of task difficulty elicit differential neural activity during these cognitive processes.
1.3 Study Objective
The objectives of the study were to determine:
1. Neural areas of activation in novice (2nd year medical students) and expert (fully certified
gastroenterologists) clinicians during clinical reasoning and decision-making tasks.
2. If there are differential neural areas of activation in clinicians with different levels of
expertise.
3. If there are differential neural areas of activation when facing straightforward (easy) or
more complex (hard) clinical cases.
1.4 Hypotheses and Predictions
Hypotheses and predictions for neural areas of activation in this study were separated into
two cognitive processes: clinical reasoning and clinical decision making.
16
1.4.1 Hypotheses related to neural areas of activation in clinical reasoning:
1. Common neural areas associated with WM will be activated in novices and experts in
both easy and hard clinical tasks because of a general interdependence of WM and
reasoning.
2. There will be greater activation of the PFC in novice participants while reading harder
cases because of increased demand on WM.
1.4.2 Predictions related to neural areas of activation in clinical decision making:
1. Novice and expert clinicians will demonstrate shared neuronal processing in some basic
form due to WM demand in clinical decision making, resulting in PFC activations
across groups and task difficulty.
2. Differential neural activity within the PFC between novice and expert clinicians will be
elicited as a result of modifiers selected in this research (level of expertise, and task
difficulty).
1.5 Thesis Outline
The cognitive processes of clinical reasoning and decision making are dependent on human
memory, as they rely on accessing previously learned information. The thesis is arranged to link
the literatures of memory, reasoning, decision making, and neuroimaging. Chapter 2 provides an
overview of background information and a literature review of clinical reasoning and decision-
making theories. In this chapter, information about the structure of human memory and how
previously learned information can be accessed in either declarative or non-declarative forms is
presented. Each of the clinical reasoning and clinical decision-making processes are then defined
followed by an overview of knowledge structure, and a literature review of associated theories
17
within medical education, the most recent of which are neuroimaging studies. Chapter 3
describes basics of functional neuroimaging, including an overview of how fMRI works,
experimental tasks and design, data acquisition, data analysis, and safety considerations.
Chapters 4 and 5 are manuscripts of the research conducted for this thesis. Tackling the two
distinct yet complimentary cognitive processes of interest, Chapter 4 presents information on the
neural areas of activation during clinical reasoning, and Chapter 5 on the neural areas of
activation during decision making. Supplementary results not included in the manuscripts are
analyzed in Chapter 6, followed by an overall discussion and conclusion in Chapter 7.
18
Chapter Two: BACKGROUND AND LITERATURE REVIEW
This chapter provides an overview of the structure of human memory important for the
retrieval of information, and outlines how previously learned information can be accessed in
either declarative or non-declarative forms. Each of the clinical reasoning and clinical decision-
making processes are defined followed by an overview of knowledge structure, and a literature
review of associated theories within medical education.
2.1 Memory Structure and Organization
A wide range of neural areas are required to be able to store, retain and recall
information,32,8. The cognitive processes of clinical reasoning and clinical decision making are
dependent on being able to access stored information, and this ability is supported by memory
structure33. This section of the thesis provides an overview of memory and associated neural
areas, as they provide the underlying framework for this research, and are of importance for
interpreting findings in upcoming chapters of this thesis.
Of critical importance to clinical reasoning is a physician’s ability to store information in
long-term memory (LTM), and eventually retrieve that information. Retrieval of information,
necessary for reasoning, language comprehension and processing28, is made possible with
working memory (WM) which allows one to ‘keep things in mind’ during complex tasks by
enabling the temporary storage and manipulation of necessary information,17,34,35. Components
of working memory include the central executive (attention-control system), visuospatial
sketchpad (maintains visual, spatial and kinesthetic information), the phonological loop (verbal
and acoustic material), and the episodic buffer (temporary interface allowing integration of
19
information from a variety of sources)17,36-38. With central executive and working memory being
reliant on the prefrontal cortex (PFC) and key for processing tasks such as reasoning, this neural
area is of interest in our work36,39. More specific localizations also of relevance include the
dorsolateral prefrontal cortex (DLPFC) for goal maintenance, executive control40, evaluating
explanations41 and selective attention42; the ventrolateral prefrontal cortex (VLPFC) for
generating explanations41 as well as simple memory recall42; and the inferior posterior parietal
cortex for phonological storage of verbal information28,43,44.
When information is stored in LTM structure, it is stored in either declarative (explicit
and analytical) or non-declarative (implicit and non-analytical) forms of memory45. Declarative
memory, characterized as being flexible22 and conscious in nature42, is further subdivided into
semantic and episodic memory45. Semantics is the memory of facts or knowledge about the
world which are not tied to any particular experience33, whereas episodic memory is
autobiographical and formed through personal experience, which embeds experiences in unique
spatial and temporal contexts46. There are fMRI studies demonstrating the semantic system to be
broadly reliant on neural areas within the frontal cortex, medial temporal lobe (MTL), and
temporal cortex8,21,47,48, and episodic memory within the PFC, inferior frontal gyrus, the
hippocampus and MTL28,49,50. Non-declarative memory on the other hand, is characteristically
unconscious, and can be tied to aspects found in the learning (encoding) phase42,22. Formed
through repetition and priming, non-declarative memory includes procedural, priming, classical
conditioning and non-associative learning, and have been respectively shown to rely on the
neocortex, striatum, cerebellum, and reflex pathways22. These forms of memory and associated
neural correlates discussed are summarized in Table 2.1.
20
Type of Memory Associated Neural Correlates Working Memory PFC36,39 Central Executive PFC36,39 Semantic Memory Frontal cortex, MTL, Temporal
cortex8,21,47,48 Episodic Memory PFC, Inferior frontal gyrus, Hippocampus,
MTL28,49,50 Phonological Storage of Verbal Information Inferior posterior parietal cortex 28,43,44 Procedural Neocortex22 Priming Striatum22 Classical Conditioning Cerebellum22 Non-Associative Learning Reflex pathways22 Table 2.1 Summary of Memory Components and Neural Correlates
Despite attempts to neatly pinpoint underlying neural areas associated to declarative and
non-declarative memory22, general consensus is that specific functional locations might not exist,
and that neural areas are widely distributed and vary depending on the nature of the task16,36,39,51.
While previous work serves as a basis for understanding neural areas implicated in different
types of memory and executive functions, there remains a lack of information about the neural
underpinnings in clinical reasoning and clinical decision making. Domain specific information
for the medical education community is needed for understanding what supports these specific
types of cognitive functions and processes more clearly, which are explored thoroughly in
chapters four and five of this thesis. As a visual summary of the memory structures discussed in
this chapter, Figure 2.1 offers a depiction memory structures supporting information retrieval,
necessary in clinical reasoning and decision-making tasks.
21
Figure 2.1 Depiction of Memory Structure Supporting Information Retrieval. Derived
from22,37,38,52,53.
2.2 Overview and Definitions of Clinical Reasoning and Clinical Decision Making
Terminology used in clinical reasoning and decision-making literature can be confusing
because of studies being conducted by different disciplines such as psychology, sociology,
philosophy, education, medicine, nursing, each portraying the topic through their own lens and
lexicon54. Terms like “decision making” and “clinical reasoning” are used interchangeably, each
with multiple surrogate terms. Proposed in a concept analysis of clinical reasoning55, surrogate
terms can be categorically used in relation to antecedent (prior to or during) and consequential
(endpoints or outcomes) of reasoning. Antecedent surrogate terms include clinical reasoning,
diagnostic reasoning, cognitive thought processes, and neurocognitive clinical reasoning. Terms
related to the outcome of these processes include clinical decision making and problem solving.
For this researach, terms will be used in relation to the step of the thought process being
discussed. Clinical reasoning will be used as the primary term when discussing the process
Working Memory
Visuospatial Sketchpad (VSSP)
Phonological Loop (PL)
Episodic Buffer
Central Executive (Temporary storage & manipulation of
information)
Speech & Language
Interface between PL & VSSP Visual &
Spatial
Long Term Memory
Declarative Memory
Non-‐ Declarative Memory
Semantic (Facts)
Episodic (Events)
Procedural Priming Classical Conditioning
Non-‐Associative Learning
22
prior to or during attempts to solve a medical problem and is defined as “a complex cognitive
process that uses formal and informal thinking strategies to gather and analyze patient
information, evaluate the significance of this information and weigh alternative action55”.
Clinical decision making will be used as the primary term when discussion relates to the
clinician’s actual execution of clinical reasoning in a final decision, path or course of action after
the benefits, consequences, and purposes of actions or sequences of actions have been
evaluated56.
2.3 Theories and Methods in Studying Clinical Reasoning and Decision Making
Early research proposed reasoning to be an all-purpose skill, which could be successfully
applied to any domain of knowledge. This turns out to be a problematic portrayal, as research
has shown that reasoning and problem solving across a wide range of problems in different
domains of knowledge produce inconsistent correlations in performance57. These findings imply
success in reasoning and problem solving are dependent on access to relevant knowledge, termed
‘content specificity’57,58. Before neuroimaging was possible, understanding how relevant
information was organized and retrieved during clinical reasoning and decision-making tasks
were based on observations of performance and through introspective or retrospective reports4,58-
60. These approaches explored stages of clinical reasoning and decision making during cue
acquisition, hypothesis generation, cue interpretation, and hypothesis evaluation1. From this,
knowledge structure representations and theoretical explanations of reasoning and decision-
making strategies have evolved.
Knowledge structure has been explained in terms of illness scripts61,62, schemas63-66, and
exemplars54,67. Retrieval of knowledge is subsequently explained by theoretical representations
23
of clinical reasoning and decision making, and include strategies such as backward reasoning
(hypothetico-deductive), forward reasoning (inductive and scheme-inductive reasoning), pattern
recognition, recognition-primed decision making, and dual processing. Further to this,
advancements in technology and use of functional neuroimaging has helped identify neural areas
important in memory, language, planning, and problem solving8,33. Preliminary studies using
fMRI within medical education have attempted to elicit neural areas supporting analytic and non
analytic reasoning, as portrayed in dual processing theories3,13. The following sections of this
chapter will now provide further overview of each of these relevant aspects in more detail.
2.4 Knowledge Structure and Expertise
The way knowledge is structured and subsequently retrieved has been shown to differ based
on content familiarity (in this work, knowledge of gastrointestinal disease), level of experience,
and demand of the task. The most common approach for studying knowledge structure is by
examining differences between novices and experts68, and it has been noted that novices use
basic science and declarative knowledge in reasoning and decision making54, whereas experts
offer more coherent explanations and show greater accuracy in reasoning and decision making
while using less basic science69. More formal structures of knowledge include illness scripts,
schemas, and exemplars (defined below). No one specific knowledge structure is thought to be
used exclusively in different levels of clinicians, meaning it is likely clinicians access any one of
these forms of knowledge structures in the retrieval information as needed54.
While there are various presentations of ways in which medical knowledge structures are
developed and how they are used in clinical reasoning and decision making, there is general
agreement that development of expertise in clinicians is due to multiple exposures to clinical
24
problems, allowing the clinician to incorporate knowledge and clinical information into pre-
existing categories54. Dispersed learning and repeated testing also appear critical in refinement of
knowledge structures and for developing accurate retrieval of relevant information from LTM
structures70,71.
Illness scripts are defined as ‘story-like narrations’ that contain information about relevant
pathophysiology, signs, symptoms, and potential diagnoses61. They compartmentalize illnesses
and diseases into prototypical categorizations62. Because of the compartmental nature of illness
scripts, additional signs and symptoms relevant to an illness or disease category can be
incorporated into pre-existing scripts, making them modifiable and accessible when similar
patterns in a clinical situation present54,72.
Schemas are mental representations that are generalizations of multiple similar clinical
problems into one structure65. Schemas allow clinicians to categorize clinical problems quickly
and make predictions regarding a course of action because information related to an illness is
structured in representations similar to flow diagrams or decision trees54. As expertise develops,
knowledge becomes ‘encapsulated’ or ‘chunked’ together, allowing more integrated information
to be organized and accessed for clinical reasoning and decision making58.
Exemplars are knowledge structures developed from clinical experiences. The
accumulation of many different clinical experiences is thought to be unconscious or non-
analytical in nature, making it difficult to clearly understand how knowledge in these forms of
structures are organized, as they are typically unconscious and therefore not available in think
aloud protocol for examination73. This unconscious, non-analytical nature of exemplars for
knowledge structures are what make them different from schemas and illness, which are more
analytical forms of knowledge structure54,65.
25
2.4.1 Novice vs. Expert Clinical Reasoning
Expert and novices differ in clinical reasoning in a couple of notable ways: 1) experts are
able to differentiate clinically relevant information when a clinical case is presented and generate
earlier more accurate hypotheses1,74, and 2) experts have more abstract knowledge
representations which allows them to encapsulate knowledge in broad ways using more general
concepts65. Novice-expert studies exploring strategies used in clinical reasoning have reported
that despite use of hypothetico-deductive approaches in both novices and experts, with no
differences in the number of hypotheses generated, content specificity of hypotheses generated
and diagnostic accuracy were greater in experts1. This suggests that along with reasoning
strategies used in clinical reasoning and decision making, underlying information contained in
knowledge structures play an important role1.
2.5 Theoretical Explanations of Clinical Reasoning and Decision Making Strategies
Retrieval of relevant information from LTM is necessary in clinical reasoning and decision
making, and this section summarizes theoretical explanations describing strategies clinicians use
to do so.
2.5.1 Hypothetico-Deductive (Backwards) Reasoning
In hypothetico-deductive reasoning, working hypotheses about the clinical diagnosis are
tested in order to confirm or reject possible solutions based on the data obtained during clinical
encounters75. This clinical reasoning strategy has been demonstrated to be used by clinicians of
all levels of expertise60, and has been shown to yield improved overall diagnostic accuracy when
compared to forward reasoning6. This point was exemplified in research done with
electrocardiogram interpretation, where groups instructed to test hypotheses (backward reason)
26
and then gather the relevant information to check if the solution seemed plausible were 50%
more accurate than when the problem was solved using forward reasoning76. The improved
accuracy was felt to be due to the fact that groups instructed to use forward reasoning gathered
more information, much of which was irrelevant to the clinical problem. Hypothesis generation
between levels of expertise have shown differences as well, in that experts tend to generate
earlier and more relevant hypotheses than novices, and produce more accurate diagnoses with
less information than is required for novices57,77.
2.5.2 Inductive and Scheme Inductive (Forward) Reasoning
Scheme inductive reasoning is an analytical technique where reasoning is guided by a
“scheme” or decision tree, in which information about clinical and relevant physiological
information is organized in a mental framework60. Forward reasoning requires the clinician to
systematically obtain clinical information in a progressive fashion, and then assign a relevant
diagnosis60. When comparing scheme-inductive, pattern recognition and hypothetico-deductive
approaches to clinical problems in gastroenterology in novices and experts using think aloud
protocol, data demonstrated pattern recognition and scheme-inductive reasoning as compared to
hypothetico-deductive reasoning were 10 and 5 times more accurate, respectively60. While
reports suggest there is a greater likelihood of diagnostic success by using schemes, others have
suggested this approach would not be sufficient for more complex problems given the lack
flexibility in schemas, as they provide only categorical representations of knowledge66.
2.5.3 Pattern Recognition and Recognition Primed Decision Making
Pattern recognition is the process where presenting signs and symptoms are compared to
previous encounters or prototypical examples57. Use of pattern recognition is thought to be non-
analytic in nature and a result of the development of well-formed, clinically relevant knowledge
27
structures based on experiential knowledge, especially with typical cases78. In recognition-
primed decision making (RPD), clinicians recognize and then assess a situation to determine
achievable goals, identify cues that are of most importance, and perform mental simulations to
decide if the context specific course of action is likely to be successful or correct79. Because of
the short the time frame in which RPD occurs, this process has been suggested to be non-
analytical in nature79. When assessing diagnostic capacity by comparing students who have been
taught using analytic techniques (such as forward or backward reasoning) versus non-analytic
techniques, diagnostic success has been comparable if not better in the group trained to use non-
analytic techniques60. This being said, reliance on non-analytic techniques alone may not work
in all circumstances, as clinical reasoning using non-analytical methods can be influenced by
first impressions, decreasing likelihood for exploration of alternative diagnoses80.
2.5.4 Dual Processing Theories
Cognitive process involved in clinical reasoning are likely interactive and iterative which,
depending upon the clinical problem, could use one or a combination of the approaches
previously described in order to arrive at clinical diagnosis81,82. The dual-processing theory is
inclusive of both non-analytic (type 1: fast, intuitive, automatic processing) and analytical
approaches (type 2: slow, deliberative, logical processing)29,83. When initially faced with a
clinical problem, it is felt that an automatic or intuitive response would be unconsciously
activated (pattern recognition), and that analytical processing would be engaged to confirm or
validate the initial hypothesis (hypothetico-deductive), especially in atypical, high-stake or
complex situations75. In attempt to make the cognitive process of clinical reasoning more
thoroughly represented, as is with the inclusion of both analytical and non-analytical components
within the dual processing theory, the dynamic interactions of when and how each of these
28
approaches are utilized have yet to be fully understood84.
2.6 Neuroimaging Studies Conducted within Medical Education
Within the limited number of initial studies using fMRI in medical education literature,
clinical reasoning and decision making have been explored in a wide variety of ways including
radiologic interpretation tasks, laparoscopic surgery training, associative learning tasks, while
cognitively operating under self-reported sleepiness, and in relationship to analytic and non-
analytic processing. This initial collection of fMRI studies highlights the complexity of this
topic, demonstrating there exist many forms of clinical reasoning and decision-making tasks
within medicine (visual, tactile or cognitive), relevant states or contexts to consider (sleepiness,
burnout), guiding theoretical frameworks (as outlined above), and structural frameworks
(knowledge and memory) to explore from.
Radiologic interpretation tasks are primarily visuospatial in nature, and two studies have
compared neural correlates of visual processing when using clinical images to non-clinical, every
day objects. Haller & Radue85investigated neural activations when radiologic and non-radiologic
images were interpreted by experts (experienced radiologists) and lay persons (non radiology
subjects). This work showed experts were significantly faster in visual interpretations and there
were two notable neural differences between experts and laypersons. One difference was
enhanced neural activations in experts when they viewed radiologic images in areas associated to
visual attention and memory retrieval. The second difference was that experts demonstrated
modified visual processing from laypersons, and relied more on neural areas associated with
mental rotation of images, attention, and spatial working memory. Melo et al.9, sought to
determine if diagnosing lesions in x-rays and naming animals embedded in x-ray images
29
produced similar patterns of neural activations, and as well introduced a simple letter naming
component to their trial as a control task. Results showed there were common areas of activation
in naming lesions, animals, and letters, but that these areas were activated to different degrees
based on the visual task. Their work suggests that medical diagnosis is a process similar to visual
categorization in every day life, but that diagnosing lesions was associated with activation in
higher order cortical areas and slower response times for interpretation.
Bahrami et al.10, investigated neuroanatomical correlates of laprascopic surgical training
in novices using simulated task trainers. This type of task is both visual and tactile in nature, and
was done during a learning (encoding) phase. fMRI results showed that the regional extent of
brain activation increased with task complexity. There were also increasing activations in the
parietal cortex, characteristically found in tasks that require more attention, and greater activation
of the medial frontal gyrus in tasks with the highest level of complexity. This was interpreted to
suggest that tasks, which are more complex in nature, involve higher executive control and
decision-making capacity.
Durning et al.86 examined how neural activations were impacted by factors such as sleep
and burnout in residents and faculty physicians. Higher self-rated sleepiness and increased burn
out was reported more in resident physicians, and was associated with decreased activation in the
DLFPC and the medial PFC. These findings are noted to be of concern given these areas are
important for error detection and working memory.
Investigation surrounding effective learning in associative learning tasks done by Downar
et. al.12 was explored with a cognitive decision-making task in which participants were instructed
to appropriately select fictional treatments for which they had no prior knowledge of, in a series
of sixty four simulated patients with acute myocardial infarction. After selecting a treatment
30
course after each patient encounter, participants were notified if their treatment was a success or
failure. Through this feedback, they were to associate when selected treatments were most
effective. Findings from this work showed a negative correlation between years of clinical
performance and number of successful treatments, that low performers learned more from
chasing success, and that high performers learned more from failures. Neural areas activated
were found to be common areas, but activated for different reasons. The right DLPFC;
important for identifying salient features, associative learning and predicting error, showed
increased activation in low performers after success, and in high performers after failure.
Bilateral inferior parietal lobules, important for attention and salience, were activated in low
performers after success, and in high performers after failure. Results from this work point out
low performers who chase success may be at risk for confirmation bias and the development of
false-beliefs, and is also notable for demonstrating distinct neural patterns in learning.
Outlined further in chapter 4 (first manuscript) of this thesis are two important studies
using fMRI discussing clinical reasoning that are non-visual in nature. The first is an fMRI
study that attempted to identify functional differences in analytic versus non-analytic reasoning
in expert clinicians3, and the second fMRI study explored non-analytic (non-declarative)
reasoning in novice and experts13. Results from the first study were interpreted to suggest that
greater activation in the prefrontal cortex is associated with analytical reasoning, based upon the
assumption that analytical reasoning was represented by incorrect answers, guessing, and deep
thought. Results from the second study suggested a common neural network between novices
and experts during non-analytical reasoning, and that experts use non-analytical reasoning more
so than novices because they demonstrated increased efficiency in the prefrontal cortex (meaning
less PFC activation).
31
As a result of this review, it is evident much work has been done to explore the supports
and processes involved in clinical reasoning and decision making with indirect methods, or by
applying research findings from other disciplines to medical education. These results have
shaped theoretical explanations to date, which can now be re examined with neuroimaging
techniques. By doing so, we can go back and address a key component currently missing from
these theoretical explanations by asking what supports these cognitive processes from a neural
basis. The few fMRI studies conducted within medical education have started to ask this
question, however, data related to this is limited at present and the approach of exploring neural
activations in clinical reasoning from a memory framework within medical education has yet to
be done. The upcoming section of this thesis introduces the neuroimaging method of fMRI,
followed by two manuscript chapters conducted for this research, which ultimately aimed to
address the question just posed.
32
Chapter Three: BASICS OF FUNCTIONAL MAGNETIC RESONANCE IMAGING
This chapter provides a brief overview of functional magnetic resonance imaging (fMRI)
technology, research design, data acquisition and analysis. This chapter does not explicitly
review all of the analysis details included in this research, as they are described in full in the
following chapters. Instead, it provides an overview of basic principles involved in this fMRI
research so upcoming chapters are more easily understood.
3.1 Basics of MRI and fMRI
Magnetic resonance imaging (MRI) images are made possible by hydrogen nuclei, found
in water molecules through out the body. Application of different radio frequency (RF) pulse
sequences within the magnetic field cause hydrogen nuclei to respond with a signal called
nuclear magnetic resonance (NMR)87-89. This signal varies based on the types of tissue found
within the body, and produces three-dimensional (3D) data called voxels87,90. This volumetric
data is represented by x (left-right dimension), y (anterior-posterior dimension), and z (inferior-
superior) coordinates, which allows for localization and comparison of data91-93.
Functional magnetic resonance imaging (fMRI) differs from standard MRI in that images
are made possible by tracking hemodynamic response to neural activity over time90,94. When
neurons become active in response to a task or demand, hemodynamic changes of increased
blood volume, increased blood flow and alterations in oxygenation occur88,95. These changes
produce the blood oxygen level dependent (BOLD) signal, which can be simplistically described
as a ratio of oxygenated to deoxygenated hemoglobin92.
Deoxyhemoglobin, being paramagnetic, is more attracted to the magnetic field and
interrupts magnetic resonance signals more so than oxyhemoglobin, which is diamagnetic, less
33
magnetically attracted, and less interruptive of the magnetic resonance signal88. These
differences in magnetic properties are what provide a natural contrast for fMRI data analysis.
The underlying assumption in fMRI is that decreased deoxyhemoglobin, and therefore increased
oxyhemoglobin concentration, indicates nearby neural activity87. Stated differently, the
observable BOLD signal, in which images are more intense when there is decreased
deoxyhemoglobin88, enables researchers to make inferences about how imposed cognitive tasks
impact neural activity92.
3.2 Tasks and Experimental Design
A block design was chosen for this research, where tasks were presented in a sequential
manner with alternating periods of stimulation and rest89. Specifically, experimental blocks of
reasoning, decision making, confidence and feedback each alternated with rest blocks called
fixation periods. Fixation periods served as baseline, and can be thought of as a control
condition during which no task is being performed96. In fixation periods, no text was presented
and no response was expected from participants; they simply looked at a plus sign (fixation
cross) on the screen.
To determine neural activities for each phase, averaged neural activity across fixation
trials are contrasted to averaged neural activity in defined experimental blocks, such as reading
(clinical reasoning phase) or answering multiple-choice questions (clinical decision-making
phase). When differences in level of activation are found to be greater during the experimental
phase of the task (i.e. during reading) than activation during fixation (baseline/rest), neural
activity is interpreted as being attributed to the cognitive process of clinical reasoning89. This
method of comparison is known as cognitive subtraction and relies on the assumption of
34
linearity, meaning neural activity is transformed into the BOLD signal in a linear fashion in
which experimental stimulus leads to increased neural activity, producing an observable BOLD
signal97.
In the block design of this research, each clinical case presented was carried out as a
‘run’. A run is defined as an uninterrupted presentation of an experimental task used in fMRI90.
Here, each run lasted for 180 seconds. Runs were presented in random order for all of the 16
different clinical cases. The timing diagram for one run of this research is presented in detail in
Figure 3.1. The top of the image depicts how long each phase of the run was presented for in
seconds (s). The bottom part of the image is an example of what participants saw on the display
screen for the phases of reading (clinical reasoning), multiple choice questions (MCQ; clinical
diagnosis), confidence (how sure the participant was in their diagnosis), and feedback (displayed
the correct answer and participant selected answer). In this research, phases of clinical reasoning
and clinical decision making were specifically targeted for analysis; analyses of confidence and
feedback phases are past the scope of this thesis and remain for exploration in future work.
Figure 3.1 Timing Diagram and Case Presentation of One Functional Run
Fixa1on 10s
Read 80s
Fixa1on 10s
MCQ 20s
Fixa1on 10s
Confidence 20s
Fixa1on 10s
Feedback 20s
35
3.3 Sample Size
Contrary to behavioural research, it is common for fMRI studies to run 12-20 participants in
an experiment90. Determining sample size in fMRI research is rarely done, and differs from
methods used in behavioural research for a few reasons98. The first reason is due to the sheer
amount of data and variability in inter-subject and intra-subject data created with this
technology99. Images are rapidly collected in fMRI, allowing for whole brain imaging in just 2-3
seconds87. A second reason is related to practical considerations such as data handling
restrictions, access to scanning time, and cost implications98,100. A third reason is that fMRI can
set out to understand neural activity in two different ways (typical or average), which impacts
how data can be used to make inferences at a population level100.
This research aimed to understand typical aspects of neural function as opposed to average
information about neural activation in clinical reasoning and decision making. The aim in
understanding typical cognitive function in this research was to identify if neural areas of
activation typically differ between novices and experts in easy and hard clinical reasoning and
decision-making processes, rather than trying to quantitatively report statistically how much
more neural activity is found in one group over another. The findings from this research provide
information on if neural areas are the same or different for specified contrasts (novice, expert,
easy, hard). For this type of inference, fixed effects analyses can be used to characterize neural
areas of activation in novices and experts on easy and hard clinical tasks at a population level,
and allow for the fact that some participants may not show this effect100.
36
3.4 Pre Processing
After initial acquisition and image reconstruction, further preparation of fMRI data before
it moves on for higher levels of analysis is required. Pre processing provides quality assurance
of the data and aims to remove unintended variability90. Pre processing options carried out within
this research using data analysis software packages are outlined in table 3.1.
Pre Processing Option Description Quality assurance Visual inspection of raw data to identify missing or cut off parts of the
brain91. Brain extraction Removal of the skull and other non brain tissue which are not of
interest in analysis91. Co-registration Alignment of participant functional data to participant structural scans
for spatial resolution92. Slice timing correction Aligning all brain slices to one reference slice because brain slices on
acquisition can be obtained in different orders. Interleaved is where odd number slices are collected first in ascending order and then even numbered slices are obtained next in descending order92.
Head motion Corrections or removal of scans with excessive head movement to prevent potential detection of erroneous activations on analysis91,92.
Normalization Registrations of the participant’s structural scan to a standardized brain. Done so that the coordinates of an identified significant cluster can be looked up in an atlas, to reduce any impacts of anatomical variability across subjects, and to align data from different participants into a common spatial framework so data can be combined for group analysis91. The MNI 152 atlas was used for this research92.
Spatial smoothing Intentional blurring of data to reduce unwanted noise and increase signal91. Done to make the distribution of the BOLD response more normal and transforms the data so it can satisfy assumptions of statistical models as well as protect against false positive findings92.
Table 3.1 Pre Processing fMRI Data
37
3.4.1 fMRI Data Analysis
The general linear model (GLM) is used for most fMRI data analyses90,101,102, and was
carried out in this research using FSL software package (FMRIB’s software Library, Version
6.0; FMRIB Analysis Group, Oxford University, UK). During analyses, basic principles of the
GLM for fMRI data are explained by the following formula:
𝑌 = 𝐺 × 𝛽 + 𝜀
The GLM attempts to find the set of experimental parameters (𝛽); (n rows [time points]
by M columns [regressors]), for a design matrix (G); (V rows [voxels] by M columns [parameter
weights]) that best explains the observed data (Y); represented by two-dimensional points; (n
rows [time points] by V columns [voxels]), by minimizing the explained error (𝜀) (n rows [time
points] by V columns [voxels])90.
A three level GLM was carried out for analyses in this research specifically for clinical
reasoning and clinical decision making. There were further aspects to the task following
diagnosis of the case presentation, however, these are beyond the scope of this thesis and are left
to be discussed in future work. In the first level of voxel-wise data analyses, every run of each
participant data was analyzed in relation to the time series of activation. Given functional runs
were fixed, meaning the order or sequence did not change, timing files specifying onset of each
stimulus within the run (i.e. onset of fixation or onset of reading) were used so that data could be
analyzed with respect to time points of interest. There were five time points of interest in each
run of the first level data; fixation (baseline), reading (clinical reasoning), MCQ (clinical
decision making), confidence, and feedback. These parameters of interest and timing of
activations are presented in Table 3.2. With onset of timing activations of interest specified,
individual (lower-level) subject data were analyzed in relation to contrasts of interest. The five
38
contrasts of interest in this work are presented in Table 3.3. Data from this level of analyses
provide mean activation levels for each participant for each of the contrasts of interest.
Events Timing Onset(s) (Running time in seconds)
Timing Duration (In seconds)
Fixation 0, 90, 120, 150 10 (for each fixation) Reading 10 80 MCQ 100 20 Confidence 130 20 Feedback 160 20 Table 3.2 First Level Parameters of Interest and Onset Times
Contrasts Reading_Fix MCQ_Fix Confidence_Fix Reading_MCQ MCQ_Reading
Table 3.3 Contrasts of Interest for First Level Participant Analyses
Second level group analyses are performed using the output of the combined estimates
from the individual (lower-level) subject analyses101. In this level, individual activation estimates
were analyzed in groupings of easy and hard questions. There were four contrasts of interest for
this level of analyses, as presented in Table 3.4.
Contrasts Easy Hard Easy_Hard Hard_Easy
Table 3.4 Contrasts of Interest for Second Level Group Analyses
In the third and final level of analyses, outputs of the combined estimates from the second
level contrasts (easy and hard) were analyzed to contrast groupings of novices and experts.
There were four contrasts of interest for this level of analyses, as presented in Table 3.5
39
Contrasts Novice Expert Novice_Expert Expert_Novice
Table 3.5 Contrasts of Interest for Final Level Analyses
3.4.2 Statistical Inference
Subject data obtained in this fMRI research are representative of each participant’s brain.
These pieces of data are called voxels, analogous to 3D pixels, and are volumetric in nature92.
Each voxel is represented by 3D coordinates (x,y,z), which are used to identify associated
structural areas using brain atlas tools. In this work, the Harvard-Oxford Cortical Structural
Atlas as well as the Harvard-Oxford Subcortical Structural Atlas within the FSL software
package was used. Typical voxel sizes are very small (3mm x 3mm x 3.5 mm92), meaning each
participant scan can produce immense amounts of data. When voxels are discussed as groupings,
they are called clusters, which represent a group of voxels in close proximity to one another.
During data analysis for this research, a subtractive approach was used to determine if the
BOLD signal was more prominent in task conditions as compared to baseline conditions89. With
this method, images obtained during tasks of interest (such as in reasoning or decision-making
phases) were compared against images obtained during baseline tasks (staring at the fixation
cross). If BOLD responses during task of interest images meet or exceed the statistical threshold
set over the BOLD responses found in baseline tasks, neural activations are attributed to the
performance of the task of interest89. A cluster threshold of p<0.05 and z = 2.5 were selected
within the FSL software package for statistical threshold. This means only groups of 50 or more
contiguous voxels (clusters) meeting the set statistical threshold would be considered areas of
activation91. Region of interest (ROI) analyses were used to investigate specific areas of the
40
brain that were hypothesized to be of importance in this work. In ROI analyses, only patterns of
activation in the specified brain region are analyzed, rather than the whole brain90,103.
3.5 Hardware and Software
fMRI research requires integration of many different types of hardware and software.
Each type used is presented in order from initial scan to final analysis in Table 3.6.
Hardware/Software Description 3T MRI Scanner • The MRI scanner has a large super conducive magnet with a
‘bore’ and a table that can move in and out of the bore. • Participants are advanced into the bore during the scanning
session, where the magnetic field is strongest89. Head Coil • A cage-like device placed over the participants’ head to further
increase the magnetic field around the area of interest; done to improve image quality and immobilize the head.
• Attachments for mirrors on the head coil allow participants’ to read rear-projected information in the bore while lying down.
Rear Projector and screen • Displays written content required for the task. Hand Held Response Pads • Given to participants so indication of response can be done
without needing to talk. • Response pads communicate to software collecting fMRI data
during scanning so response times and data timings are related. Presentation Software • Required to display content required for the research task.
• Must be linked to the scanning software so that both systems operate precisely together.
Scanning Software • Software required for obtaining anatomical and functional images.
Data Storage Server and Software
• Separate servers and software for data storage is used to keep obtained data obtained from initial scanning.
Data Analysis Software • Required for the analysis of functional neuroimaging data. Computers with a large amount of processing power are required for these software packages to run.
Post Analysis Data Presentation Software
• Separate software for creating images of results.
Table 3.6 Required Hardware and Software for fMRI Research
41
3.6 Participant Considerations
3.6.1 MRI Safety
Two main points of emphasis related to MRI safety include ensuring participants are safe
to enter into the magnet, and protecting participants from loud acoustic noises produced by the
scanner. MRI screening forms, found in Appendix A, are used to ensure participants have no
metal implants or devices prior to entering the magnetic field. Acoustic noises produced by MRI
scanners when images are obtained and can be as loud as 140dB. Comparable to standing within
100 feet of a jet engine (asha.org), this level of noise will damage unprotected ears104. To protect
participants, earplugs as well as noise cancelling headphones are required to be worn during
scanning.
3.6.2 Exclusion Criteria
Important exclusion criteria outside those listed on the MRI screening forms are found in
Appendix B, and aim to ensure participants have healthy normal brains which have not been
altered by trauma, neurovascular changes or disease processes. Medications or substances that
alter cerebral function or cerebral blood flow were also screened for given the importance of
hemodynamic integrity when using the BOLD signal.
3.6.3 Practical Considerations
Further to the safety concerns and exclusion criteria mentioned, practical considerations
included asking participants if they wore glasses or suffered from claustrophobia. Metal on
glasses frames can be an easily overlooked detail. This can be navigated by asking participants
to either wear contact lenses for the scan, or to let the researcher know the prescription of the
glasses lenses so an MRI safe set of goggles could be used. If the participant is aware of having
claustrophobia, further discussion about their ability to tolerate one hour in the bore of the
42
magnet is important to have prior to study enrolment. The bore of the magnet is rather narrow,
which may cause discomfort to persons who suffer from fear of small spaces. In all scans,
participants are made aware that they can stop at any time by pushing a squeeze ball in the
scanner. Medications to reduce anxiety were not an option for this study, and potential
participants who had concern about tolerating the scan were thanked for their interest, but
ultimately not recruited.
43
Chapter Four: MANUSCRIPT # 1
4.0 Making up your mind: neural areas of activation during clinical reasoning
Pam Hruska, Olav Krigolson, Sylvain Coderre, Kevin McLaughlin, Filomeno Cortese,
Christopher Doig, Tanya Beran, Bruce Wright, Kent Hecker
Formatted for manuscript-based thesis. Has been submitted to Medical Education.
4.1 Abstract
Purpose
Clinical reasoning is dependent upon working memory. More precisely, during the
clinical reasoning process stored information within long-term memory (LTM) is brought into
working memory (WM) to facilitate the internal deliberation that affords a clinician the ability to
reason through a case. In the present study, we examined the relationship between clinical
reasoning and working memory while participants read and diagnosed clinical cases with
functional magnetic resonance imaging (fMRI). More specifically, we examined the impact of
clinical case difficulty (easy, hard) and clinician level of expertise (2nd year medical students,
senior gastroenterologists) on neural activity within regions of cortex associated with working
memory (i.e., the prefrontal cortex) during the clinical reasoning process.
Method
fMRI was used to scan ten second-year medical students and ten practicing
gastroenterologists while they reasoned through sixteen clinical cases (eight straight forward
[easy] and eight complex [hard]) during a single one-hour scanning session. Within-group
analyses contrasted the easy and hard cases (easy > hard, hard > easy) which were then
44
subsequently utilized for a between-group analysis to examine effects of expertise (novice >
expert, expert > novice).
Results
Reading clinical cases evoked multiple neural activations in occipital, prefrontal, parietal,
and temporal cortical regions in both groups. Importantly, increased activation in the prefrontal
cortex in novices for both easy and hard clinical cases suggests novices utilize WM more so than
experts during clinical reasoning.
Conclusion
This study demonstrates there is an important relationship between clinical reasoning and
human working memory. We found that clinician level of expertise elicited differential
activation of regions of the human prefrontal cortex associated with WM during clinical
reasoning. As such, we suggest future models of clinical reasoning take into account that use of
working memory is not consistent through out all clinical reasoning tasks, and that this memory
structure may be utilized differently based level of expertise.
4.2 Introduction
Effective and safe patient care depends on sound clinical reasoning and diagnosis29-30,
therefore proficiency in both reasoning and decision making are important abilities for a
physician1. Reasoning and decision-making stages are distinct from one another in that clinical
reasoning is the activity prior to or during attempts to solve a medical problem, whereby a
clinician weighs and sorts through assessment details obtained from medical history, physical
assessment and test results. Subsequent to this, clinical decision making is the stage during
which the clinician chooses between competing options or courses of action to assign a final
diagnosis and determine the plan of care55. The processes of reasoning and decision making are
45
complimentary to one another and require integration of information from basic science, medical
knowledge and clinical experience54. In this study, we focus on clinical reasoning and how this
cognitive process is supported by long term and working memory.
4.2.1 Memory and Reasoning
Though memory and reasoning have been considered separate topics in literature and
contemplated from insulated experimental paradigms and theoretical models, recent work
exploring many different types of tasks demonstrate a relationship between these cognitive
activities15,14. Relationships between memory and reasoning are noted in dual-task demands,
tasks requiring manipulation of memory content, and tasks where memory content must be
coordinated for integration into a new domain15,24. Parallels to each of these mentioned tasks are
found in clinical reasoning and as such, it can be implied memory processes and clinical
reasoning are interdependent14,15,105. Indeed, a physician’s ability to access stored information
from long-term memory (LTM) is of critical importance so that appropriate knowledge
schemas58 and/or illness scripts62,64 can be mobilized for evaluation of the presenting medical
problem75. The need to access LTM during clinical reasoning implicates working memory (WM)
as being crucial to the reasoning process. Specifically, in order for a physician to use LTM
during reasoning the relevant LTMs need to be brought forward into WM17,24,34,35 where they can
be accessed and manipulated in order to clinically reason and diagnose clinical cases. In other
words, the clinical reasoning process is dependent upon and sub-served by WM.
Studies using functional magnetic resonance imaging (fMRI) have demonstrated that the
neural locus of WM is prefrontal regions of the human cortex16,106,107. The prefrontal cortex
(PFC) is thought to be where the central executive resides, which is the high level system that
controls WM and, thus, the retrieval and access of LTM36,39. More specific localizations that may
46
also be relevant include the dorsolateral prefrontal cortex (DLPFC) for goal maintenance,
executive control40 and selective attention42, as well as the ventrolateral prefrontal cortex
(VLPFC) for attention control and simple memory recall – purportedly in the format of the
phonological loop and the visuospatial sketchpad37,42.
While previous neuroimaging research serves as a basis for understanding neural areas
implicated in WM in reasoning tasks from other realms, there is need for further information
about the neural underpinnings in clinical reasoning and the role of WM in this process.
4.2.2 fMRI and Medical Education
fMRI studies conducted within medical education have mostly investigated the decision-
making phase9,11,12 or are visuo-spatial in nature10,85. There are, however, two relatable pieces of
work to date discussing clinical reasoning that are non-visual in nature. The first is an fMRI
study that attempted to identify functional differences in analytic versus non-analytic reasoning.
fMRI images obtained in this work were done during three contrasting phases; reading, which
was treated as baseline neural activation, answering, and reflecting. Immediately after scanning,
participants engaged in a think aloud protocol where they were asked to describe the process
they used to answer each question. Results were interpreted to suggest that greater activation in
the PFC is associated with analytical reasoning, based on the assumption that analytical
reasoning was represented by incorrect answers, guessing, and deep thought.
The second fMRI study explored non-analytic (non-declarative) reasoning by ten internal
medicine interns (novices) and seventeen board-certified staff internists (experts)13. In phase one
of the experimental task, the multiple-choice question (MCQ) was presented during a reading
phase. In phase two, answer options were presented for participants to select from, followed by
phase three in which participants were instructed to silently reflect on how they arrived at their
47
diagnosis. Results from this work suggested a common neural network between novices and
experts during non-analytical reasoning, and that experts had increased efficiency (decreased
activation) in the PFC. This prefrontal efficiency was interpreted as increased use of non-
analytical reasoning processes in experts83.
4.3 Purpose
Here we present the first fMRI study focusing on differences between novice and experts
during clinical reasoning to highlight the role of working memory in this cognitive process. The
specific aim of the present study was to explore neural areas of activation in novice (2nd year
medial students) and expert (senior gastroenterologists) clinicians during clinical reasoning tasks
and to see whether areas of activation differed when cases were straightforward (easy) or more
complex (hard). Our hypotheses were twofold: (1) Common neural areas associated with WM
would be activated in novices and experts, with both easy and difficult clinical scenarios because
of a general network demonstrating interdependence of WM and reasoning and (2) there would
be greater activation of the PFC in novice participants while reading harder cases because of
increased demands on WM.
4.4 Methods
4.4.1 Participants
Twenty healthy volunteers with normal or corrected-to-normal vision completed the
present experiment in full. Ten second year medical students (8 male, mean (range) age 26.5
(22-38) years, SD = 5.3) were the novice participant group and ten currently practicing
gastroenterologists (5 male, mean (range) age 39.5 (32-50) years, SD = 4.5) were the expert
clinicians. Novice participants were all from the Cumming School of Medicine at the University
of Calgary and had completed the gastrointestinal course one year prior to this research. Expert
48
participants were all currently practicing gastroenterologists with formal academic teaching
responsibilities at the Cumming School of Medicine at the University of Calgary and within
Alberta Health Services in Calgary, Alberta, Canada. We restricted participation to right-handed
individuals due to predominant language processing being lateralized to the left cerebral
cortex108,109. Exclusion criteria included inability to complete an fMRI due to scanning safety
risks (metal, anxiety, claustrophobia, pregnancy), medical history of traumatic brain injury,
history of non-medically induced loss of consciousness for >10 minutes, psychiatric illness
requiring medical treatment, past or present use of psychotropic drugs, history of seizures, any
other diagnosis of neurovascular or neurophysiologic abnormality, or use of calcium channel
blockers. Participants were free to withdraw from the study at any time. This study was
conducted in accordance with the ethical standards outlined by the Calgary Health Ethics
Research Board (CHREB), and the Seaman Family MR Research Center at Foothills Medical
Centre.
4.4.2 Stimuli and Procedures
Participants read sixteen gastroenterology clinical cases during a single one-hour fMRI
scanning session. Each case was approximately 215-219 words of written text in length, and was
shown via a mirror on a rear-projected screen situated above and behind the participant’s head
via the fMRI scanner projection system (Avotec Inc, Florida, USA).
4.4.3 Clinical Cases
Clinical cases were optimized for display from previously published written cases for this
research74. Eight of the clinical cases were deliberately made to be “easy” and eight were made
to be “hard”. For easy questions, the patient's initial contextual data was concordant with the
analytical data subsequently presented. For hard questions, the patient's contextual data was
49
discordant with the analytical data subsequently presented. Questions were randomized during
scans. There were four questions related to each of the following clinical presentations; elevated
liver enzymes, diarrhea, dysphagia, and anemia.
The experiment was performed using Presentation® (Version 16, www.neurobs.com).
Each case presentation began with a centrally presented fixation cross for 10 s after which a
randomly selected case was presented for participants to read for 80 s. There were other aspects
to this experiment, such as a clinical decision-making component, and these components are
presented in subsequent papers. This is the first of two studies during which we focused solely
on the clinical reasoning process.
4.5 Data Acquisition and Analysis
4.5.1 Functional and Structural Data Acquisition
Data were acquired on a 3-Tesla GE Discovery MR750 diagnostic magnetic resonance
whole body scanner (General Electric Healthcare, Waukesha, WI, USA) at the University of
Calgary, Seaman Family MR Research Centre at Foothills Medical Centre. Preliminary imaging
consisted of acquiring a T1-weighted 2D spin-echo sequence with the same geometric
orientation and voxel size as the subsequent functional images. The functional imaging
sequences were gradient-recalled echo, echo-planar imaging (GRE-EPI) sequences in the
oblique/axial plane and were acquired in an interleaved, bottom-up slice acquisition (repetition
time [TR] = 2000 ms, echo time [TE] = 20 ms, flip angle [FA] = 70°, 37 slices at 3 mm
thickness, in-plane resolution of 64 x 64 pixels reconstructed in a field of view [FOV] of 24 cm)
using a 16-channel receive-only phased array head coil. Each functional run began with 6 TRs
during which no data were acquired to allow for steady-state tissue magnetization. A total of 90
50
echo-planar imaging volumes were collected in each functional run, and a total of 16 functional
runs were collected for each participant. A 3D high-resolution (1 x 1 x 1 mm), T1-weighted axial
image was also obtained from each participant (FOV = 25.6 cm) for registration of the functional
data.
4.5.2 fMRI Data Processing and Analysis
Data were pre processed using FEAT (Version 6.00), which is part of FSL (FMRIB’s
Software Library, Version 6.0; FMRIB Analysis Group, Oxford University, UK). Data were
motion corrected110, registered by FLIRT111, and spatially smoothed with a Gaussian kernel of
5.0 mm full width at half maximum. The resulting time series was then convolved using a
gamma function.
For the first-level analysis, each case for all 16 functional scans for every participant was
separately analyzed contrasting reading (the first 20 s of the 80 s reading phase) with the first
fixation phase (10 s; read > fixation). The parameter estimate maps and variance maps for the
contrast were subsequently forwarded to a second-level fixed-effects analysis where all
functional scans (16) for each participant were combined into easy (8) and hard (8) average
images. We combined contrasts for the easy and hard clinical cases at this point, and also
contrasted easy relative to hard cases (easy > hard) and hard relative to easy cases (hard > easy).
All runs were given equal weighting in the model.
A final third-level mixed-effects analysis using FLAME 1 + 2 (due to the small number
of participants in the present experiment (http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM) was
conducted where data from the second-level contrasts of the two groups of participants were
combined separately to model group level differences and to contrast group effects. Thus, at this
final stage of analysis we had contrasts reflecting the group effects separately (novice vs. expert)
51
and contrasts examining group difference (novice > expert, expert > novice).
Statistically significant clusters of activation were initially identified on the entire group
statistical map by using a voxel-wise threshold to z > 2.3 (p < 0.05) and the FSL cluster analysis.
However, given our outlined hypotheses, we also conducted specific region of interest (ROI)
analyses103. In this analysis, ROIs were defined within the prefrontal cortex (DLPFC, VLPFC)
and similar statistical criteria were used to evaluate activation: a voxel-wise threshold to z > 2.3
(p < 0.05) and a criteria of at least 50 contiguous voxels112.
We used the Montreal Neurological Institute (MNI) coordinates of the voxel within the
cluster with the maximum z statistic to determine the most probable anatomical label for the
cluster from the Harvard-Oxford Cortical Structural Atlas packaged in FSL. We have also
included the reading versus fixation contrast for both easy and hard clinical cases for both groups
to show whole brain activation.
4.6 Results
4.6.1 fMRI
Our initial analyses were focused on separate examinations of the group activation maps
for the novice and expert clinicians while they read the easy and hard clinical cases. By doing
this, we hoped to identify regions of interest (i.e., differential activation between easy and hard
cases and/or novices and experts) for subsequent analyses. As expected, reading clinical cases
(both easy and hard) evoked significant changes in hemodynamic activity in multiple brain
regions for both novice and expert clinicians – see Figure 4.1 and Tables 4.1 and 4.2 for full
details. We conducted follow up analyses where we contrasted the two groups (novice > expert,
expert > novice) that included underlying contrasts that compared the reading of the easy and
52
hard clinical cases (easy > hard, hard > easy) in addition to group contrasts of the easy and hard
cases in isolation. These analyses revealed that there were greater activations in the left temporal
pole and anterior division of the left middle temporal gyrus for novices relative to experts when
reading both easy and hard clinical cases (see Figure 4.2 and Table 4.3). In other words,
activation in the left anterior temporal lobe was greater for novices than for experts – but was the
same for both the reading of easy and hard clinical cases.
In a final series of analyses, we conducted specific ROI analyses to examine activity in
the PFC (specifically DLPFC and VLPFC). The results of these analyses (see Figure 4.2 and
Table 4.4) revealed that for novices, reading more difficult clinical cases resulted in greater
prefrontal activation than reading easy clinical cases, and further, this activation was greater in
novices than in experts.
4.7 Discussion
The relationship between clinical reasoning and working memory was explored using
fMRI. The significant findings from this study were: (1) There were common neural areas of
activity in both novice and experts while reasoning through clinical cases, including the right
middle frontal gyrus, right pre central gyrus, left frontal pole, right frontal pole, left frontal
orbital cortex, and left superior frontal gyrus; (2) There were greater activations in the left
anterior temporal lobe for novices relative to experts when reading both easy and hard clinical
cases; and (3) There was increased prefrontal activity for hard questions in novices relative to
experts, and this increase in activation is also relative to easy questions. Stated differently,
significantly more working memory support was required for novices than for experts, especially
as they reasoned through hard clinical cases.
53
As hypothesized, there exist common neural areas of activations in novice and expert
clinicians associated with working memory when reasoning through clinical cases. Common
prefrontal activations found in the present study are consistent with areas identified in a meta-
analysis of 189 neuroimaging studies related to WM24. As such, this reinforces our initial
expectations of working memory being critical to clinical reasoning. As a result of finding
shared neural areas of activation across all levels of expertise and task difficulties, one could
infer that activation of the PFC and use of WM is an index, or measure of, clinical reasoning.
Differences in PFC activity between novice and experts, in which novices demonstrated
increased neural activation for both easy and hard clinical cases, could also suggest novices
utilize WM more so than experts. More significant prefrontal activations in novices could be
explained by the importance of this region in WM in guiding semantic memory retrieval to
reason through scenarios, especially during complex clinical cases in which there are more
competing clinical distractors28,36. When specifically considering increased right frontal polar
areas of activations in novices, other work has suggested this area is important in episodic
memory tasks requiring ongoing monitoring during retrieval113. As well, while some have
suggested decreased prefrontal activity as a hallmark of expertise13, an alternative view is that
neural areas of activation shift or functionally reorganize as expertise and knowledge structures
develop85,114. Our findings of significantly increased prefrontal activations in novices could,
therefore, be related to WM guiding semantic retrieval, ongoing monitoring during episodic
retrieval, or due to functional reorganization in expertise.
We have demonstrated instances when neural areas diverge based on task complexity or
clinician level of expertise. By using more drastic contrasts in level of expertise and question
difficulty than in previous work, our research findings support the comment that in the trajectory
54
of development to expertise, there may be phases of neural patterns exhibited in clinical
reasoning13. We have also identified that reading clinical cases, considered the reasoning phase
in our work, produces distinct demands on neural activity, and support previously noted concerns
that reading might not serve as the best baseline contrast for fMRI research in medical
education3.
4.7.1 Other Areas of Activation
Activation of the left anterior temporal lobe in novices for both easy and hard clinical cases
suggests this group relies more heavily on LTM during clinical reasoning. The anterior temporal
lobe is associated with human conceptual knowledge, and more specifically the meaning of
words and objects across many domains115,116. This finding is less frequent in fMRI research,
which more commonly attributes semantic processing to the medial temporal lobe, but supports
the notion that anterior temporal lobes have greater recruitment when more precise recall of
information is required in semantic tasks115. This point has been explored in language network
neuroimaging studies, where the anterior temporal lobe is noted to be important in text
comprehension and for creating coherent representations of dialogue or information23,26,45.
Consequently, semantic processing can be understood as important for language processing and
as well for accessing knowledge in clinical reasoning processes21. There have also been accounts
of the anterior temporal lobe being activated during the retrieval of abstract concepts, specifically
related to words, which are required for judgments28. Given the type of information presented in
clinical cases is more familiar and less abstract to experts, the anterior temporal lobes may not
have shown significant recruitment as less effort for creating coherent representations of
information is needed because the knowledge presented is more general or common place to that
55
level of clinician.
4.8 Limitations
There are limitations to this study. First, there was a non-random selection of participants
in this study due to limited availability of experts to draw from. Second, focusing specifically on
clinical cases related to gastrointestinal illness tests only one area of clinical knowledge, which
may not be representative of how one reasons through other clinical aspects of knowledge related
to other body systems or clinical topics. Third, information presented had no associated images,
and so results cannot be generalized to clinical reasoning tasks that are visuo-spatial in nature.
4.9 Conclusion
Our work demonstrates the role and importance of WM to clinical reasoning. Clinician
level of expertise elicits differences in neural areas activated in clinical reasoning tasks, and
demonstrates that novice clinicians rely more heavily on WM than experts, especially during
hard tasks. Importantly, our research provides a more direct understanding of neural areas
activated in clinical reasoning. Though our research design explored a specific type and context
of medical knowledge, these contributions can improve our understanding of how novices and
experts access and use WM during clinical reasoning. Continuing to partition phases of clinical
reasoning and decision making in future fMRI research may also be of importance within
medical education, as results to date offer different perspectives on what constitutes a reasoning
phase, and when reasoning and decision making are actually occurring. It is plausible reasoning
and decision making are happening at different points of time for novices and experts, and
disentangling each of these processes as well as clarifying what constitutes a reasoning task in
56
clinical education might allow for more thoughtful future investigations.
Figure 4.1 Combined Neural Areas of Activation in Clinical Reasoning
Top row (Right hemisphere & Left hemisphere): novice brain. Bottom row (Right hemisphere &
Left hemisphere): expert brain. Reading Easy Cases (blue) vs. Reading Hard Cases (red).
Common areas of activation (purple).
57
Figure 4.2 Region of Interest Images in Clinical Reasoning
Novice > Expert, Easy and Hard Cases (green). Novice>Expert (red); Hard > Easy (blue).
58
Hemisphere Region
Cluster
Size Max Z MNI
Coordinates X Y Z Reading Easy Cases * Right Occipital Fusiform Gyrus 22695 6.04 36 23 25 * Right Middle Frontal Gyrus 533 3.97 20 80 51 Right Precentral Gyrus 256 4.92 27 61 68 Right Juxtapositional Lobule Cortex 233 3.58 42 66 59 Left Frontal Pole 195 3.86 57 94 32 Right Frontal Pole 157 4.2 27 88 46 Right Heschel’s Gyrus (Includes H1+H2) 155 3.58 20 55 39 Right Postcentral Gyrus 84 3.49 11 58 52 Left Frontal Orbital Cortex 80 4.03 62 79 24 Right Paracingulate Gyrus 71 3.42 39 72 57 Left Superior Frontal Gyrus 53 4.16 51 71 70 Right Frontal Pole 53 3.59 28 90 29 Reading Hard Cases * Right Lateral Occipital Cortex, Superior
Division 25279 7.04 35 23 61
* Right Frontal Pole 1038 4.96 21 83 40 Right Frontal Pole 365 4.28 26 90 31 Left Paracingulate Gyrus 337 3.62 47 67 60 Right Frontal Pole 211 5.29 25 88 43 Right Middle Temporal Gyrus, Anterior
Division 182 3.79 15 65 25
Right Superior Temporal Gyrus, Posterior Division
178 3.73 11 55 37
Left Superior Frontal Gyrus 68 5.51 48 66 72 Right Postcentral Gyrus 66 3.33 12 58 49 Right Temporal Pole 54 3.28 22 67 26 * Cluster list of activations Table 4.1 Reading: Common Clusters for Novices Reading Easy and Hard Clinical Cases
59
Hemisphere Region Cluster Size
Max Z MNI Coordinates
X Y Z Reading Easy Cases * Left Inferior Temporal Gyrus,
Temporoocciptal Part 12538 5.14 69 33 28
* Left Middle Frontal Gyrus 4736 4.95 70 79 48 * Right Middle Frontal Gyrus 1079 4.49 18 75 53 Left Superior Temporal Gyrus,
Posterior Division 311 3.74 70 49 34
Left Paracingulate Gyrus 272 3.51 48 81 54 Left Frontal Pole 184 3.67 58 96 37 Right Frontal Pole 169 4.65 35 97 40 Right Precentral Gyrus 148 3.8 24 65 47 Right Insular Cortex 77 3.26 30 74 34 Right Frontal Pole 60 3.37 42 93 50 Reading Hard Cases * Left Inferior Temporal Gyrus,
Temporooccipital Part 18308 5.23 69 39 23
* Left Superior Frontal Gyrus 8800 5.19 59 77 64 Left Frontal Pole 132 3.92 59 92 46 Right Precentral Gyrus 123 3.54 29 62 55 Left Middle Temporal Gyrus, Anterior
Division 122 3.62 74 61 27
Right Frontal Pole 99 4.05 39 98 31 Right Frontal Pole 69 3.49 36 96 43 Right Frontal Pole 69 3.34 19 83 35 Right Frontal Pole 62 3.74 31 96 39 Left Frontal Orbital Cortex 58 3.6 58 80 24 * Cluster list of activations Table 4.2 Reading: Common Clusters for Experts Reading Easy and Hard Clinical Cases
60
Hemisphere Region Cluster Size
Max Z MNI Coordinates
X Y Z Easy Cases Left Temporal Pole 472 4.15 -56 16 -8 Hard Cases Left Middle Temporal Gyrus, Anterior
Division 560 5.28 -58 -2 -18
Table 4.3 Reading: Novice > Expert Activations
Hemisphere Region Cluster Size
Max Z MNI Coordinates
X Y Z Right Superior Frontal Gyrus 183 2.49 35 79 56 Right Frontal Pole 92 3.24 21 82 38 Right Superior Frontal Gyrus 49 2.87 43 84 58 Right Frontal Pole 41 3.47 28 86 29 Right Frontal Pole 38 2.27 22 83 43 Left Frontal Pole 38 3.18 64 89 38 Table 4.4 Reading: Areas of Activation in Regions of Interest: Novice > Expert; Hard >
Easy
61
Chapter Five: MANUSCRIPT # 2
5.0 Mind Made Up: Neural areas of activation during clinical decision making
Pam Hruska, Kent Hecker, Sylvain Coderre, Kevin McLaughlin, Filomeno Cortese, Christopher
Doig, Tanya Beran, Bruce Wright, Olav Krigolson.
Formatted for manuscript based thesis.
5.1 Abstract
Purpose
In the second of a two part series, this paper builds from findings that working memory supports
clinical reasoning and suggests this memory structure is also essential in clinical decision
making. With effective and safe patient care dependent on a clinician’s ability to accurately
diagnose, this paper focuses on understanding what modifies clinical decision-making output and
how working memory is utilized within this form of cognitive processing.
Method
Neural performance during clinical decision making was captured using functional magnetic
resonance imaging (fMRI) to specifically determine areas of activation in novice and expert
clinicians and to see if these changed as a result of level of expertise and/or task difficulty.
fMRI data was collected from ten second year medical students (novices) and ten practicing
gastroenterologists (experts) while they diagnosed sixteen (eight easy and eight hard) clinical
cases using multiple-choice questions (MCQ).
62
Results
Participants were more accurate on easy than on hard cases (p < 0.02), and answered hard cases
more slowly than easy cases (p < 0.001). Experts correctly diagnosed more cases than novices
and made their diagnoses faster than novices (p < 0.005, respectively). fMRI data demonstrated
that overall, regions of activation during clinical decision making on easy cases were not
significantly different between novice and expert clinicians. However, when assigning diagnoses
to hard MCQs, novice clinicians had greater activations in the left anterior temporal cortex and
left ventral lateral prefrontal cortex. Conversely, expert clinicians had greater activations in the
right dorsal lateral, right ventral lateral, and right parietal cortex. Support from WM was vital in
clinical decision making across all levels of expertise and task difficulties; however, distinctions
and hemispheric differences were noted.
Conclusions
In our work, we have helped establish that the use of working memory (WM) and
activation of the prefrontal cortex (PFC) is a common neural area recruited by novice and expert
clinicians’ when diagnosing easy and hard clinical cases. We have also elicited neural
distinctions between novice and expert clinicians and as well during variation in clinical task
complexity. While both novices and experts demand use of the PFC, there are hemispheric
activation differences, as well as differences in recruitment of other supportive brain regions
between the levels of clinicians. During the development of expertise and progression of case
difficulty, our data could suggest WM and supporting areas of the PFC evolve from use of
semantic, factual knowledge that is rule-based guided by basic causal explanations, to processes
that are guided by more internal experiences, which allow for comparison between exemplars by
dedicating more attention for evaluative assessment.
63
5.2 Introduction
Clinical decision making is the clinicians’ actual execution of clinical reasoning into a
final decision, path or course of action55. Often carried out within high-stake situations117, it
requires integration of information from basic science, medical knowledge and clinical
experience54. In the second of a two part series, this paper builds from previous findings which
suggest working memory (WM) supports clinical reasoning, and aims to understand the ways in
which this memory structure supports clinical decision making. With effective and safe patient
care dependent on a clinician’s ability to accurately diagnose29,30, it is of interest to determine
what modifies clinical decision-making output as well as understand how WM is utilized within
this form of cognitive processing. Neural performance during clinical decision making was
captured with functional magnetic resonance imaging (fMRI) to specifically determine areas of
activation in novice and expert clinicians and to see if these change as a result of level of
expertise and task difficulty.
5.2.1 Clinical Decision Making: Stages and Categories
Clinical decision making has been explained in two overlapping ways; by theories that
have explored stages of decision making and by theories which have categorized decision
processes based on analytical or non-analytical characteristics4-118 . Understanding performance
during different stages of clinical reasoning and decision making since Elstein’s1 proposed four-
stage model (cue acquisition, hypothesis generation, cue interpretation, and hypothesis
evaluation) has evolved over years of research. Major themes find experts, when compared to
novices, generate faster and more accurate solutions in problem solving, detect patterns, analyze
problems more qualitatively, are better able to self-monitor for mistakes, solve problems with
forward reasoning, take advantage of any and all sources of information available, all while
64
expending less cognitive effort73. Analytical forms of decision making include, but are not
limited to, hypothetico-deductive (backward) reasoning and inductive and scheme-inductive
(forward) reasoning6, where as non-analytical approaches include pattern recognition57, and
recognition-primed decision making119.
Dual processing theories (DPT), prevalent in medical education and decision-making
literature, attempt to explain how and when analytical and non-analytical types of decision
processes are utilized and exist within one’s mind 120,121,83. Non-analytical processes, referred to
as type 1 decision making, are characterized as being autonomous, fast, intuitive, automatic, and
non-declarative, whereas analytical, or type 2 processing is regarded as being slow, deliberative,
logical, and declarative5,83. Having been presented in diverse ways, DPT literature is moving
away from previous held beliefs that decision making is dependent on two distinct systems, each
with their own underlying and dedicated neural areas121, and instead towards the idea that one
mind that elicits forms of neural distinction which oscillate dependent on modifiers which have
yet to be fully defined84,83. After consideration and debate of these theories, Evans and
Stanovich83 suggest type 2 processing supports hypothetical thinking and loads heavily onto
WM. Given WM is supported by the PFC, it is of particular interest to determine how this neural
area performs in our work.
5.2.2 fMRI of Clinical Decision Making
There are few preliminary studies looking at clinical decision making within medical
education using fMRI3,12,13,122. Decision making in radiology and laparoscopic surgery have
mainly addressed visual-spatial decision-making systems and dexterity tasks123-125. Notable
findings from this work are that medical diagnosis tasks are more cognitively demanding than
non-medical ones, requiring increased recruitment of higher order prefrontal and cingulate
65
cortices9. This provides validation for the need of domain specific research, and suggests
training may lead to moved or increased neuronal activations rather than decreased neuronal
activity in experts85.
Two important studies related to decision-making research using written and verbal tasks
within medical education have been conducted. Durning et al.3 hypothesized functional
differences in analytic versus non-analytic reasoning of expert internists. In this work,
participants answered a series of higher order multiple-choice questions using fMRI3. Neural
images were obtained during three contrasting phases; reading (treated as baseline neural
activation for analysis), answering, and reflecting. Immediately after scanning, a think aloud
protocol was used to categorize participant responses based on correct vs. incorrect, guess vs.
non-guess, and deep vs. superficial answers. Guessing, incorrect answers, and deep thought were
interpreted to be indicative of analytic reasoning. These identified behavioural categories were
used to subsequently guide fMRI data analysis to determine neural areas associated with analytic
reasoning. Findings showed greater activation of the prefrontal cortex with answering
incorrectly along with significantly less activation in the bilateral precuneus and left mid
temporal gyrus when participants were guessing3.
The second and most recent piece of work explored the neural basis of non-analytical
reasoning in novices (internal medicine interns) and experts (attending physicians with faculty
appointments). Findings suggested shared neural areas for decision making between the two
groups and identified experts had greater neural efficiency in the prefrontal cortex when
compared to novices13. The underlying assumption in this research design was that by moving
away from use of the think aloud protocol, a technique requiring participants to ‘talk through’ the
mental process they used to make a decision4, non-analytical thinking in participants would be
66
captured. Evans and Stanovich83 suggest type 2 (analytical) processes are dependent on working
memory as a defining feature over type 1 (non-analytical) processing; therefore, the PFC could
be a defining neural area for analytical reasoning.
5.2.3 Neural Areas of Interest
The predominant neural region of interest, which supports working memory and is
involved in decision making is the prefrontal cortex (PFC)20. Different subdivisions of the PFC
also noted to be engaged in decision making in humans in general include the dorsolateral
prefrontal cortex (DLPFC) and the frontopolar cortex126. The PFC is linked to WM, which is
responsible for retrieving information from long-term memory (LTM) to be used in reasoning,
language comprehension, planning, assessment, execution and outcome processing24-26,27,28. The
DLPFC is involved in maintaining and manipulating information in WM, and the frontal polar
cortex implicated in rule-based deciding126.
Based on the previous discussions and findings from relatable fMRI work, we made two
predictions: (1) Among novice and expert clinicians, there would be evidence of shared
neuronal processing in some basic form due to WM demand in clinical decision making,
resulting in PFC activations across groups and tasks and (2) novice and expert clinicians would
elicit neural areas of distinction as a result of manipulations to task difficulty, highlighting
subdivisions of the PFC which oscillate84 dependent on our modifiers selected (level of expertise,
and task difficulty).
67
5.3 Methods
5.3.1 Participants
This study was conducted in accordance with the ethical standards prescribed in the
Declaration of Helsinki, the Calgary Health Ethics Research Board (CHREB), and the Seaman
Family MR Research Center. Exclusion criteria included inability to complete an fMRI due to
scanning safety risks (metal, anxiety, claustrophobia, pregnancy), medical history of traumatic
brain injury, history of non-medically induced loss of consciousness for >10 minutes, psychiatric
illness requiring medical treatment, past or present use of psychotropic drugs, history of seizures,
any other diagnosis of neurovascular or neurophysiologic abnormality, or use of calcium channel
blockers. We restricted participation to right-handed individuals due to predominant language
processing being lateralized to the left cerebral cortex109.
Twenty healthy volunteers with normal or corrected-to-normal vision completed the
present experiment in full. Ten second year medical students (8 male, mean (range) age 26.5
(22-38) years, SD = 5.3) were the novice participant group and ten currently practicing
gastroenterologists (5 male, mean (range) age 39.5 (32-50) years, SD = 4.5) were the expert
clinicians. Novice participants were all from the Cumming School of Medicine at the University
of Calgary and had completed the gastrointestinal course one year prior to this research. Expert
participants were all currently practicing gastroenterologists with formal academic teaching
responsibilities at the Cumming School of Medicine at the University of Calgary and within
Alberta Health Services in Calgary, Alberta, Canada.
5.3.2 Stimuli and Procedures
During scanning, participants read sixteen clinical cases during a single one-hour fMRI
session. Each case was approximately 215-219 words of written text in length, and was shown
68
via a mirror on a rear-projected screen situated above and behind the participant’s head on the
fMRI scanner projection system (Avotec, Inc., Florida, USA). For each case, participants were
given eighty seconds to read a clinical scenario and then were asked to indicate the single most
likely diagnosis, presented in the form of a multiple-choice question with four answer choices.
Participants answered using a MR-compatible response box (Cedrus, San Pedro, CA, USA)
positioned by their hands and had twenty seconds to make their selection. Use of fixation periods
in-between reading and answering tasks for a length of ten seconds was used to establish
cognitive baseline for contrast during data analysis.
5.3.3 Clinical Cases
The experiment was performed using Presentation® (Version 16, www.neurobs.com).
Clinical cases were optimized for display from previously published clinical cases to fit the
screen projection area74. Eight of the clinical cases were deliberately made to be “easy” and eight
were made to be “hard”. For easy questions, the patient's initial contextual data were concordant
with the analytical data subsequently presented. For hard questions, the patient's contextual data
were discordant with the analytical data subsequently presented. Lab values and case
presentations contained information in text form only; no graphic images were presented.
Questions were randomized during scans. There were four questions related to each of the
following clinical presentations; elevated liver enzymes, diarrhea, dysphagia, and anemia.
5.4 Data Acquisition and Analysis
5.4.1 Behavioural Data and Analysis
For each case, participants’ accuracy (correct, incorrect) and response time (s) were
recorded. Overall accuracy (%) and mean response time (s) were analyzed post-experiment via a
69
2 (experience: novice, expert) by 2 (case difficulty: easy, hard) two-way analysis of variance
(ANOVA). An alpha level of 0.05 was assumed for this test.
5.4.2 Functional and Structural Data Acquisition
Data were acquired on a 3 Tesla GE Discovery MR750 diagnostic magnetic resonance
whole body scanner (General Electric Healthcare, Waukesha, WI, USA) at the University of
Calgary, Seaman Family MR Research Centre. Preliminary imaging consisted of acquiring a T1-
weighted 2D spin-echo sequence with the same geometric orientation and voxel size as the
subsequent functional images. A functional scan was performed for each clinical case trial (16
functional scans in total). The functional imaging sequences were gradient-recalled echo, echo-
planar imaging (GRE-EPI) sequences in the oblique/axial plane and were acquired in an
interleaved, bottom-up slice acquisition (repetition time [TR] = 2000 ms, echo time [TE] = 20
ms, flip angle [FA] = 70°, 37 slices at 3 mm thickness, in-plane resolution of 64 x 64 pixels
reconstructed in a field of view [FOV] of 24 cm) using a 16-channel receive-only phased array
head coil. Each functional run began with 6 TRs during which no data were acquired to allow
for steady-state tissue magnetization. A total of 90 echo-planar imaging volumes were collected
in each functional run, and a total of 16 functional runs were collected for each participant. A 3D
high-resolution (1 x 1 x 1 mm), T1-weighted axial images were also taken of each participant
(FOV = 25.6 cm) for registration of the functional data.
5.4.3 fMRI Data Processing and Analysis
Data were pre processed using FEAT (Version 6.00), which is part of FSL (FMRIB’s
Software Library, Version 6.0; FMRIB Analysis Group, Oxford University, UK). Data were
motion corrected110, registered by FLIRT127, and spatially smoothed with a Gaussian kernel of
5.0 mm full width at half maximum. The resulting time series was then convolved using a
70
gamma function.
For the first-level analysis, every functional run for each participant were separately
analyzed contrasting decision making (the MCQ phase) with the fixation phase (10 s; MCQ >
fixation). The parameter estimate maps and variance maps for the contrast were subsequently
forwarded to a second-level fixed-effects analysis where the functional scans for each participant
(16) were combined into easy (8) and hard (8) average images. We combined contrasts for the
easy and hard clinical cases at this point, and also contrasted easy relative to hard cases (easy >
hard) and hard relative to easy cases (hard > easy). All runs were given equal weighting in the
model.
A final third-level mixed-effects analysis using FLAME 1 + 2 (due to the small number
of participants in the present experiment; http://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM) was
conducted where the second-level contrasts of the two groups of participants were combined
separately to model group level differences and to contrast the group effects. Thus, at this final
stage of analysis we had contrasts reflecting the group effects separately (novice vs. expert) and
contrasts examining group difference (novice > expert, expert > novice).
Statistically significant clusters of activation were initially identified on the entire group
statistical map by using a voxel-wise threshold to z > 2.3 (p < 0.05) and the FSL cluster analysis.
However, given our hypotheses outline above, we also conducted specific region of interest
analyses (ROI)103. In this analysis, ROIs were defined within the PFC (DLPFC, VLPFC) and
similar statistical criteria were used to evaluate activation: a voxel-wise threshold to z > 2.3 (p <
0.05) and a criteria of at least 50 contiguous voxels112,128.
We used the Montreal Neurological Institute (MNI) coordinates of the voxel within the
cluster with the maximum z statistic to determine the most probable anatomical label for the
71
cluster from the Harvard-Oxford Cortical Structural Atlas packaged in FSL. We also included
the decision making versus fixation contrasts for both easy and hard clinical cases for both
groups to show whole brain activation.
5.5 Results
5.5.1 Behavioral Results
The two-way ANOVAs examining the impact of expertise (novice, expert) and case
difficulty (easy, hard) on accuracy and response time demonstrated two key results. One, experts
(78 +/- 12%) correctly diagnosed more cases than novices (58 +/- 25%), F(1,36) = 11.2, p <
0.005. Two, experts (6.0 +/- 1.8 s) made their diagnoses faster than novices (8.7 +/- 2.8 s),
F(1,36) = 17.4, p < 0.005. Additionally, participants were more accurate on easy (76 +/- 16%)
than on hard cases (60 +/- 24%), F(1,36) = 7.3, p < 0.05. Participants also answered harder cases
more slowly (8.4 +/- 2.1 s) than easy cases (6.2 +/- 2.7 s), F(1,36) = 11.2, p < 0.005. In both
analyses there were no interaction effect between expertise and case difficulty.
5.5.2 fMRI Results
Initially we focused on separate analyses to generate group activation maps for the novice
and expert clinicians while they answered the easy and hard multiple-choice scenarios. In line
with accepted practice103 we did this to identify regions of interest (i.e., regions with differential
activation between easy and hard cases and/or novices and experts) that were activated during
clinical decision making for subsequent contrast analyses. Not surprisingly, processing and
answering the multiple choice questions (MCQ; both easy and hard) evoked significant changes
in hemodynamic activity in multiple brain regions for both novice and expert clinicians – see
72
Figure 5.1 and Tables 5.1 and 5.2 for full details. In general, we saw increased activations in the
occipital, parietal, temporal, and prefrontal cortices during the answering of the MCQs. In easy
cases for both novices and experts, common neural areas were activated in the left lateral
occipital cortex, left paracingulate gyrus, and right middle frontal gyrus. In hard cases for both
novices and experts, common areas of activation were found in the left inferior frontal gyrus, left
lateral occipital cortex, right inferior frontal gyrus and left frontal pole.
In line with our predictions and supported by our observation of the group activation
maps we conducted follow up analyses where we examined novice – expert differences (i.e.,
novice > expert, expert > novice) during clinical decision making for the easy and hard cases.
We found no significant differences in brain activity between groups during clinical decision
making for the easy cases. We suggest this is because both novice and expert clinicians require
WM support during clinical decision making, as evidenced by recruitment of the PFC across
tasks. There are however, important neural distinctions, for hard MCQs. In tasks with increased
difficulty, we found that novice clinicians had greater activations than expert clinicians in the left
anterior temporal cortex and left VLPFC (see Figure 5.2 and Table 5.3). Conversely, expert
clinicians had greater activations than novice clinicians in the right DLPFC, the right VLPFC,
and the right parietal cortex (see Figure 5.2 and Table 5.4). While both novices and experts
demand use of the PFC, there are hemispheric activation differences, as well as differences in
recruitment of other supportive brain regions between the levels of clinicians.
5.6 Discussion
There were two significant findings from this study. First, fMRI data revealed no significant
differences in regions of activation between novice and expert clinicians during clinical decision
73
making on easy questions. Second, complex tasks produced hemispheric and neural activation
differences between groups, with novices having greater activation of the left anterior temporal
cortex and left VLPFC, and experts having greater activation of the right DLPFC, right VLPFC,
and right parietal cortex.
The combination of fMRI data and behavioural results suggest that even though there are
shared neural areas used by novice and expert clinicians’, increased task difficulty produces
distinct neural and hemispheric differences between the two groups during complex clinical
tasks. Experts correctly diagnosed significantly more cases than novices, and made their
diagnoses significantly faster than novices (p < 0.005, respectively). Participants were
significantly more accurate on easy than on hard cases (p < 0.01), and answered harder cases
significantly slower than easy cases (p < 0.001). These behavioural findings are consistent with
previous research related to expertise91.
No significant differences in regions of activation between novice and expert groups during
easy clinical decision-making tasks suggest commonality for how clinicians solve
straightforward problems. Within medical education fMRI literature, others have also found
similar results and suggest neural differences only become prominent in tasks that are more
cognitively demanding123. This is highlighted by Melo et al.9, where they found that radiologists
activated similar distributed neural areas when generating diagnostic hypotheses as when naming
animals and letters based on visual recognition, but noted the degree of activations differed
significantly based on cognitive demand with more of the higher order cortical areas being
evoked during diagnosis tasks.
In hard clinical cases, novices elicited significant activation in the left anterior temporal
cortex. This area has been identified as being crucial in semantic ability, which is an aspect of
74
long-term declarative memory composed of knowledge acquired about the world such as facts,
concepts and beliefs28,115. The combination of activations in the left anterior temporal cortex and
left VLPFC in novices is in line with previous literature noting that interactions between these
areas are vital for successful memory107 and important for retrieving semantic representations28.
More specifically, the VLPFC supports semantic selection in the generation of basic causal
explanations and inferences41. From this, we can imply novices solve complex clinical tasks with
the support of the VLPFC to generate basic explanations and inferences retrieved from semantic
memory.
Experts compared to novices had greater activations in the VLPFC, albeit in the opposite
hemisphere, along with involvement of the DLPFC during the hard questions. This pattern of
brain activity suggests that in addition to generating basic explanations, the DLPFC supports
evaluation of the options in light of some held normative standard and test for attributions of
correctness41. Further yet, the engagement of the parietal cortex has been noted when increased
attention is required125. In complex cases with discordant information to sort through, experts
dedicate more cognitive attentional resources relative to novices, while for easy cases recruit
areas that evaluate basic explanations.
To address the hemispheric differences between groups, it has recently been suggested the
right hemisphere supports abstract or holistic premises and the left hemisphere dependent on
concrete material129,130. Applied within the context of our study, where novices activated the left
hemisphere and experts the right, it can be suggested that novices make clinical decisions based
on concrete representations and experts with more abstract representations. Stated differently, it
is likely that novices use more concrete representations of semantic knowledge (facts) during
decision making28.
75
Hemispheric differences are also important for the discussion about how WM is differentially
utilized within decision making. With WM being dependent on the PFC, work by Krawczyk126
discussing sub processes and associated structures of the PFC is particularly helpful. The left
hemisphere (significantly recruited by novices during hard tasks), has been demonstrated as
being used when analyzing responses to environmental stimuli presented, in high similarity
decision making, for rule based processes, and when using explicit cues for guiding decision
making126. The left PFC has also been described as an ‘interpreter’ because of its role in making
sense of information and in generating causal explanations of confronting information131. Right
hemispheric activations on the other hand (significantly recruited by experts on hard questions),
have been noted to require prior knowledge and activated when making decisions amongst
multiple options (induction); in comparison between exemplars; when decisions are guided
internally by choices or based on memory and personal experiences; during ambiguity resolution
independent of explicit rules; and during option assessment and categorization or framing of
information126,130.
Our data demonstrate that decision making requires significant recruitment of the PFC in
both novice and expert clinicians, especially during more difficult tasks. These findings are
similar to work done with radiologic image interpretation, which demonstrated enhanced brain
activations in radiologists (experts) over non-radiologists (lay persons) 85. Results such as these
conflict with findings of decreased prefrontal activity (termed prefrontal efficiency) in experts in
other work13. Differences could be related to experimental design or novice-expert comparison
differences, and future research to explore reasons for discrepancy may be warranted.
How do our findings relate back to the medical decision-making literature and Dual Process
Theories? Analytical decisions (type 2) have been tied to the PFC126, suggesting that more
76
complex problems require a search of LTM and, thus, a need to access/utilize WM; associated
with greater PFC neural activation. As discussed, we found greater PFC hemispheric activation
differences for difficult questions between novice and experts suggesting the use of an analytical
decision-making process. However, we cannot provide evidence for the proposal by Evans and
Stanovich83 that type 2 (analytical) processes are dependent on working memory as a defining
feature over type 1 (non-analytical) processing. This is a methodological problem. We
determined that there were no significant differences in activation between novice and experts
for the easy questions; however, this does not mean that the PFC was dormant during these tasks
(they were active, as shown in Figure 1). There were simply no significant differences in
activation between novice and experts. Therefore, while we provide baseline information for
novice/expert neural correlates of decision making, more work on robust fMRI research designs
are required in order to pick up on differences in analytical or non-analytical processes.
5.7 Limitations
Establishing ‘baseline’ in fMRI research is complicated. Our use of fixation crosses as a
contrast as well as Durning’s3 use of reading as a contrast offer two variations in methodology,
making comparison of studies a challenge as these nuances in design affect data interpretation
and subsequent results. Future research could include the use of simple opposing tasks as
baseline27. Doing so could provide increased awareness of the effects baseline tasks have on
data analysis, and could also helpful as a method to determine if analytical or non-analytical
processes can be more clearly targeted.
Despite standardized scenarios and MCQs being persistently used in medical education,
moving away from this artificially imposed context may be important for determining how
77
clinicians truly make decision in clinical environments117. With participants being aware of test
like scenarios, increased use of analytical thought by cognitive override could be a confounding
concern132.
5.8 Conclusion
Modelling neural correlates of clinical decision making over an increased variety of
tasks, as has been done in working memory literature24, is needed within medical education to
identify whether there are common neural underpinnings in clinical decision making, and to
determine if certain tasks cause unique neural patterns of activation. In our work, we have
helped establish that the use of WM and activation of the PFC is a common neural area recruited
by novice and expert clinicians when diagnosing easy and hard clinical cases. We have also
determined neural distinctions occur within clinical decision making as a result of clinician level
of expertise and during more clinically complex tasks. This suggests WM in clinical decision
making can be utilized differently dependent on the task or level of clinician. During the
development of expertise and progression of case difficulty, our data could suggest WM and
supporting areas of the PFC evolve from use of semantic, factual knowledge that is rule-based
guided by basic causal explanations, to processes that are guided by more internal experiences,
which allow for comparison between exemplars by dedicating more attention for evaluative
assessment.
78
Hemisphere Region Cluster Size
Max Z MNI Coordinates
X Y Z Easy MCQ * Left Occipital Pole 11554 5.16 53 16 30 * Left Frontal Pole 3879 4.68 66 85 38 * Left Lateral Occipital Cortex 1209 5 52 29 67 Left Paracingulate Gyrus 474 4.13 47 82 51 Right Inferior Frontal Gyrus 402 4.85 17 74 50 Left Frontal Pole 333 4.26 51 97 31 Right Frontal Orbital Cortex 184 3.61 28 77 33 Right Hippocampus 146 3.49 32 48 33 Right Frontal Pole 87 3.77 36 96 31 Right Lateral Occipital Cortex, Superior
Division 73
3.58 26 34 58 Right Lateral Occipital Cortex, Superior
Division 68
3.69 30 30 60 Right Middle Frontal Gyrus 57 3.73 26 63 65 Hard MCQ * Right Lateral Occipital Cortex 14060 5.68 57 17 29 * Left Inferior Frontal Gyrus 6264 4.98 70 73 47 * Left Lateral Occipital Cortex 2134 4.72 50 26 65 * Right Cingulate Gyrus, Anterior Division 1310 4.99 41 78 49 Right Frontal Orbital Cortex 341 4.34 30 79 35 Right Inferior Frontal Gyrus, Pars
Opercularis 331 5.83 20 72 49 Left Frontal Pole 224 4.29 51 97 29 Right Lateral Occipital Cortex, Superior
Division 196 4.38 31 32 59 Left Frontal Pole 122 4.31 62 92 40 Left Postcentral Gyrus 120 3.42 76 52 55 Right Postcentral Gyrus 79 3.41 21 50 66 Right Frontal Pole 62 4.69 27 92 40 Right Frontal Pole 55 3.83 28 89 28 * Cluster list of activations Table 5.1 MCQ: Common Clusters for Novices Diagnosing Easy and Hard Clinical Cases
79
Hemisphere Region Cluster Size
Max Z MNI Coordinates
X Y Z Easy MCQ * Left Inferior Frontal Gyrus 12820 5.4 68 71 47 * Left Lateral Occipital Cortex 1708 4.5 58 27 63 * Right Lateral Occipital Cortex 1011 4.43 31 34 60 * Right Middle Frontal Gyrus 535 4.41 19 76 52 * Right Parahippocampal Gyrus, Posterior
Division 411 3.92 34 48 32 Right Inferior Temporal Gyrus,
Temporooccipital Part 127 3.67 14 42 28 Left Paracingulate Gyrus 108 4.26 46 80 56 Right Middle Frontal Gyrus 93 3.43 28 68 67 Right Frontal Pole 81 3.55 41 92 52 Right Frontal Pole 48 3.42 30 83 59 Hard MCQ * Left Inferoir Temporal Gyrus,
Temporooccipital Part 11407 5.04 70 34 23 * Left Inferior Frontal Gyrus, Pars
Opercularis 3947 4.51 70 73 48 * Left Lateral Occipital Gyrus, Superior
Divison 2047 4.59 58 28 63 * Right Superior Frontal Gyrus 1957 4.57 44 83 57 * Right Superior Parietal Gyrus 1141 4.13 32 37 60 Right Inferior Frontal Gyrus, Pars
Opercularis 295 3.76 30 72 47 Right Middle Frontal Gyrus 224 4.39 28 66 66 Left Frontal Pole 124 3.91 54 87 53 Right Middle Frontal Gyrus 77 3.36 29 62 62 Right Frontal Pole 68 4.41 20 85 30 Right Frontal Pole 61 3.69 28 94 34 Left Cingulate Gyrus 56 3.38 45 64 47 Left Frontal Pole 50 3.42 48 96 41 * Cluster List of activations Table 5.2 MCQ: Common Clusters for Experts Diagnosing Easy and Hard MCQ
Hemisphere Region Cluster Size
Max Z MNI Coordinates
X Y Z Left Frontal Pole 778 4.24 -52 36 -2 Table 5.3 MCQ: Hard Questions: Novice > Expert
80
Hemisphere Region Cluster Size
Max Z MNI Coordinates
X Y Z Right Frontal Pole 907 4.69 36 36 48 Right Angular Gyrus 453 4.17 44 -56 38 Table 5.4 MCQ: Hard Questions: Expert > Novice
Figure 5.1 Combined Neural Areas of Activation in Clinical Decision Making
MCQ Easy (Blue); MCQ Hard (Red); Common Areas of Activation (Purple)
Top: Right & Left hemispheres Novice brain. Bottom: Right & Left hemispheres Expert brain.
81
Figure 5.2 Novice Expert Differences in Clinical Decision Making on Hard Tasks
Hard MCQ: Novice > Expert (Blue); Expert > Novice (Red)
82
Chapter Six: SUPPLEMENTAL RESULTS
6.1 Total Study Recruitment
A total of twenty-six participants were recruited and enrolled for participation in this
research. Six full sets of participant data were excluded for analyses due to reasons such as
scanner protocol changes (one participant), scanner malfunction (two participants), participant
discomfort such as nausea and neck pain (two participants), and extreme head movement (one
participant).
6.2 Pre Screening Questionnaire Data
A pre screening questionnaire was given to every participant in this study prior to entering
the MRI room. An example this form is found in Appendix B. Information gathered using this
form is summarized in this section.
6.2.1 Education Achievement Levels
Novice participants were predominantly baccalaureate prepared, with the exception of
one participant who was pursuing their Master’s degree, and another participant who had
previously achieved a doctor of chiropractic degree. All expert participants were medical
doctors with advanced medical specialization. On top of this, one expert had achieved a
Master’s level degree, and two participants had a Doctor of Philosophy degree.
6.2.2 Formal Teaching
None of the novices were required to teach in any formal way in their current activities,
whereas all expert participants had an expectation of formal teaching within their current role.
This information is of interest, as it has been noted in literature that teaching is a form of
deliberate practice, which can reduce decline in clinician performance133.
83
6.2.3 Years of Clinical Experience
All of the novice participants were in their second year of medical school at the time this
study was conducted. In an effort to maintain anonymity of participants in the expert group,
categories for years of clinical experience were captured. Five experts had 1-10 years of clinical
experience, meaning they had 1-10 years of independent clinical practice after being done
medical school, residency, and fellowship in their areas of specialization. The other five experts
had 11-20 years of independent clinical experience. None of the data used during analyses were
from clinicians over 70 years of age, which has been associated to decrease in cognitive function
and speed of information processing20,78.
6.2.4 Hours of Sleep
There were no significant differences in the amount of hours slept between novices and
experts on the night before participation in this research (p = 0.23). Novices had on average slept
7.45 hours the night before participation in this research, whereas experts had on average 7.05
hours of sleep. This information is relevant because in other research conducted within medical
education, lack of sleep has been noted to impact cognitive performance during decision
making11.
6.3 Post Screening Questionnaire Data
After completion of the scanning session, a post-screening questionnaire to ask for
feedback on how the session went was collected from each participant. An example of this form
is found in Appendix C. 17/20 (85%) participants reported having enough time to read the
questions in full; 3/20 wished they had a bit more time to re-read parts of the clinical case. 11/20
(55%) felt comfortable in the scanner, and 9/20 (45%) had discomfort in the form of neck pain,
84
back pain, headache, and nausea. For those with discomfort during the scan, onset was felt
around half way through their scanning session, suggesting that immobilization for that period of
time may be too long for any future research. 1/20 (0.05%) participant felt claustrophobic, but
managed to complete the session in full. 9/20 (45%) participants commented that the edges of
the screen for reading were either dark or blurry, but felt they were still able to complete the
scanning session despite some distortion to clarity on the edges of the screen. All participants
felt they could hear instructions over the intercom and communicate effectively at any point
needed while in the scanner. All felt the hand boxes for responding were easy to use after
practice.
6.4 Confidence in Decisions
After each MCQ was answered, participants were asked to rate how confident they were
in their assigned diagnoses. Participants indicated their confidence rating for each case using
hand boxes and corresponding confidence levels. Table 6.1 provides a summary of how
confidence ratings were coded and captured in this study.
Table 6.1 Confidence Answer Options and Associated Ratings and Coding
Answer Options Associated Confidence Ratings
Associated Numeric Values Captured on Log Reports
A 75-100% 1 B 50-74% 2 C 25-49% 3 D 0-24% 4
85
6.4.1 Overall Confidence in Decisions
Novices had an average confidence rating of 2.05 across all 16 clinical cases, meaning on
average, they felt 50-74% confident in their diagnoses. Experts had an average confidence rating
of 1.41 across all clinical cases, meaning on average, they felt 75-100% confident in their
diagnoses. A two-way independent analysis of variance (ANOVA) examining the impact of
expertise (novice, expert) and case difficulty (easy, hard) on confidence ratings demonstrated a
significant interaction effect F (1, 316)=5.23, p<0.05, meaning the effect of task difficulty lead to
differences in confidence ratings in novices than for experts.
6.4.2 Confidence on Easy and Hard Questions
Experts were significantly more confident (M=1.09, SD 0.40) than novices (M=1.93, SD
0.91) on easy cases, t (158)=7.54, p<0.01.
Experts were also significantly more confident (M=1.73, SD 0.09) than novices (M=2.18,
SD 0.09) on hard cases, t (158)=3.52, p=0.01.
86
Chapter Seven: OVERALL DISCUSSION AND CONCLUSIONS
7.1 Key Findings
The aim of this thesis was to explore the neural areas of activation during the cognitive
processes of clinical reasoning and decision making. More direct assessments of memory
structures used and neural areas activated in clinical reasoning and decision making were made
possible by using fMRI as a tool. The main findings were as follows:
7.1.1 Behavioural Findings:
• Experts were significantly faster and more accurate in assigning clinical diagnoses than
novices on both easy and hard cases.
• Novice and expert participants were significantly faster and more accurate on easy than on
hard clinical cases.
7.1.2 Overall Neural Findings:
• There were shared neural areas of activation identified in clinical reasoning in both levels of
clinicians and tasks, including the right middle frontal gyrus, right pre central gyrus, left
frontal pole, right frontal pole, left frontal orbital cortex, and left superior frontal gyrus.
• There were shared neural areas of activation identified in clinical decision making in both
levels of clinicians and tasks. In easy cases for both novices and experts, common neural
areas were activated in the left lateral occipital cortex, left paracingulate gyrus, and right
middle frontal gyrus. In hard cases for both novices and experts, common areas of activation
were found in the left inferior frontal gyrus, left lateral occipital cortex, right inferior frontal
gyrus and left frontal pole.
87
• Neural areas of activation in clinical reasoning and decision making diverge in novices and
experts with manipulations to task difficulty as described in 7.1.3 and 7.1.4.
• Given the overall neural activity observed within the prefrontal cortex (PFC) in both clinical
reasoning and decision making, findings from this research highlight the dependence on
working memory (WM) in these cognitive processes.
7.1.3 Neural Areas of Activation and use of Memory in Clinical Reasoning
• Activation in the left anterior temporal lobe were greater for novices than for experts, but was
activated similarly in both the reading of easy and hard clinical cases in novices.
• Interpretations of these results suggest more WM support was required for novices than for
experts, especially as they reasoned through hard clinical cases.
7.1.4 Neural Areas of Activation and use of Memory in Clinical Decision Making
• Neural areas of activation during decision making on easy cases did not differ significantly
between novice and expert clinicians.
• There were hemispheric differences in activation on hard clinical cases, which suggests
clinicians of different levels of expertise employ different decision-making processes when
diagnosing hard clinical cases. Explanations offered for these differences include novices
using the left hemisphere for more concrete rule based processes, and experts using the right
hemisphere for more holistic and abstract processes for assigning diagnoses.
7.2 Discussion
Our results demonstrate examples of both common and divergent neural areas of activation
during clinical reasoning and decision-making tasks. This work adds to the literature by
providing awareness of neural areas supporting these cognitive processes and as well, exposes
88
factors (level of expertise and task difficulty) associated with differential neural activity. This
demonstrates clinical reasoning and decision-making processes are dynamic, and neural
activations may change or reorganize as experience develops or as situations demand. We have
yet to fully understand these dynamics; it will be of great interest to see how neuroscientists and
medical educationalists probe this topic further.
The beginning of chapter 2 provided an overview of human memory and explored how
previously learned information could be accessed in either declarative or non-declarative forms.
Findings from this research demonstrated novices used structures associated with concrete
factual information and semantic memory, whereas experts activated the hemisphere more
commonly associated with abstract representations.
Neural activity within the PFC in clinical reasoning and in both levels of clinicians and
task difficulties suggest these cognitive processes depend on WM. In clinical reasoning, novices
demonstrated significantly more activation in the PFC and dependence on WM than experts, but
there was still evidence demonstrating that experts, when clinically reasoning, utilized the PFC
as well. Further to this, both levels of clinicians, when diagnosing difficult clinical cases,
recruited the PFC in significantly different ways. There are two major pieces of work previously
discussed in this thesis implicated by these findings. First, Evans & Stanovich83 suggested that
within dual processing theory, type 1 processes do not require WM. Our results demonstrated
WM was required in all tasks, and as a result it is difficult to determine if the use of WM is truly
a defining hallmark between type 1 and type 2 processes. Secondly, Durning13 suggested PFC
efficiency (or decreased PFC activations) found in experts was attributed to their use of non-
analytical reasoning. The reasons our work differs from these two presented perspectives could
be related to the research design in this study, in that tasks may not be tailored enough to elicit
89
non-analytical thinking. Another reason could be that participants were aware they were being
tested in general (though not aware of the specific content area being tested on), and therefore
were primed to use type 2 or analytical thought processes. To more confidently elicit non-
analytical thought in clinical reasoning and decision-making tasks, decreasing the available time
for reading and responding, imposing constraints or blinding participants about the study may
help to explore this area further.
Concept analyses related to clinical reasoning and clinical decision making have provided
discrete definitions for each of these cognitive processes55. Results from this work support
continued use of these categorizations, as distinct neural patterns of activity were elicited in each
of the cognitive processes examined here. When separating reading (reasoning) from multiple
choice questions (MCQ; decision making) by forced fixation periods in between, it is possible
the study design used in this research contributed to these neural differences. Another
consideration in relation to the MCQ phase of this study design is that the act of deliberating and
selecting diagnoses from a predetermined list of options could contribute to distinct neural
patterns not found when clinicians are forced to independently generate diagnoses. These points
may be methodological considerations for future research, and alterations to design could
consider allowing clinicians to progress through reading (reasoning) at their own pace, moving
immediately into stating the most likely clinical diagnoses while in the scanner. Potential
barriers for approaching the design this way would be jaw movement impacting image quality
when clinicians state a diagnosis, and being able to clearly hear the clinicians answer through the
intercom system.
While we did not directly explore our findings in relation to the dual processing theory,
parallels have been drawn between declarative and non-declarative memory structures to type 1
90
(automatic/unconscious) and type 2 processing (deliberative/conscious) respectively83, meaning
results from this research could also be relevant in this realm of literature. Dual processing
theory has been examined from the perspectives of many different disciplines5. As a result,
descriptions of dual processing have been termed as ‘types’, ‘systems,’ and ‘modes,’83. With
findings in this research demonstrating common neural areas of activation in clinical reasoning
and decision making, we can confidently support proposals suggesting it is best to term dual
processing as ‘types’ or ‘modes’ of thinking, rather than ‘systems’ of thinking83. By describing
dual processing as ‘types’ or ‘modes’ of thinking, shared neural areas of activation supporting
cognitive processing in different ways is permissible. Describing dual processing as ‘systems’
implies different neural areas support different types of cognitive processing, and results from
this research do not substantiate this perspective.
From the limited collection of neuroimaging research conducted within medical
education to date, it has been identified that there exists great variation in the ways clinical
reasoning and decision making have been studied. Refinement of study design protocols are
needed, and as a result of this work, we have added some suggestions for what could be
improved upon in future investigations. Summarized, suggestions have included: continued
manipulations to baseline tasks; study design changes to impose increased constraint, decreased
time, or blinding of participants to determine if this helps further elicit non-analytical reasoning;
and potentially moving away from use of MCQs as options for assessing diagnostic ability.
7.3 Strengths
There are four primary strengths of this thesis. First, separation of clinical reasoning and
decision-making processes in the study design allow for more specific assessment of the each of
91
the cognitive processes investigated. Second, and consistent with other approaches in literature,
we contrasted novice expert clinicians. Third, we introduced different levels of task difficulties.
Lastly, and perhaps most importantly for the success of this research, we engaged a
multidisciplinary team of investigators who provided varied perspectives of expertise in medical
education, clinical practice, principles of cognition, neuroscience, and neuroimaging.
7.4 Limitations
fMRI as a method has important limitations to note, which include the BOLD response being
a surrogate marker of neural activity and representative of neuronal mass activity134. There are
also limitations to this pilot study given research constraints and design. These include: (1) a
non-random selection of participants, (2) focusing specifically on clinical cases related to
gastrointestinal illness, (3) assessing clinical reasoning and decision making in tasks which are
only reading based in nature, (4) using periods of fixation as an ideal baseline in fMRI design,
(5) inability to replicate real-life/naturalistic clinical environments within fMRI study design, (6)
participants being aware of being in a test-like situation, and (7) sampling students from a
university that teaches to schema based learning. Further to this, use of MCQs as a measure for
diagnostic ability, though commonly done for assessment of knowledge in medical education,
may force analytic reasoning and decision-making strategies. As a result, findings from this
research might not be representative of non-analytic processes. Another limitation of using
MCQs could also include answer options triggering recognition-primed decision making. A
final limitation to note is that the theoretical perspectives fMRI data are explored from could
influence interpretations. Awareness of potential researcher biases and prevention of over
92
claiming what data reveals is important to assess for.
7.5 Future Studies
We have only scratched the surface of being able to understand the neural activations and
dynamics in clinical reasoning and decision making. We are therefore at the beginning of new
paradigm in clinical reasoning and decision-making research, in which we are combining the use
of neuroimaging technics to more fully elucidate these cognitive processes. Continued use of
neuroimaging research within medical education will provide a more thorough awareness of
neural structures supporting these processes, including when and why they are engaged, and in
turn will inform theoretical representations. Ways to more thoroughly understand the neural
supports in clinical reasoning and decision making include using neuroimaging research to assess
forms of these processes (visual, tactile or cognitive), during varied relevant mental states or
environmental contexts (sleepiness, burnout, during uncertainty, within clinical practice
environments), and in relation to components of memory (encoding, storage, and retrieval).
Other considerations could include cross-institutional studies to assess if results from one
geographic location are similar or different to findings from other geographic areas.
Longitudinal research designs would be of benefit for identifying neural changes as development
of expertise occurs. Lastly, neuroimaging in this work used fMRI, and use of other
neuroimaging modalities such as electroencephalogram (EEG), in isolation or combined with
fMRI, would allow for assessment of not only the neural areas of activation, but the temporal
sequence with which they are activated, providing a more dynamic understanding of neural
activity in these cognitive processes.
93
7.6 Conclusions
Overall, this research demonstrated that fMRI is a sensitive enough tool to detect neural
areas of activation in clinical reasoning and decision making, and that two factors associated to
differential neural activity include level of expertise and task difficulty. This work informs
medical educationalists more precisely of the neural areas supporting clinical reasoning or
decision making and the role of WM used in these cognitive processes. To draw inspiration from
work done in artificial intelligence by Kurzweil31, the closer we can get to being able to precisely
understand the brain in clinical reasoning and decision making, the more able our tailoring to
impact this area of clinical education and in turn performance of clinicians.
94
REFERNCES
1. Elstein A, Shulman LS, Sprafka SA. Medical Problem Solving: an Analysis of Clinical Reasoning. Cambridge, Massachusetts: Harvard University Press; 1978.
2. McLaughlin K, Mandin H. Using “concept sorting” to study learning processes and outcomes. Academic Medicine. 2002;77(8):831-836.
3. Durning SJ, Graner J, Artino AR, et al. Using functional neuroimaging combined with a think-aloud protocol to explore clinical reasoning expertise in internal medicine. Mil Med. 2012;177(9 Suppl):72-78.
4. Banning M. The think aloud approach as an educational tool to develop and assess clinical reasoning in undergraduate students. Nurse Educ Today. 2008;28(1):8-14.
5. Evans J. Dual-Processing Accounts of Reasoning, Judgment, and Social Cognition. Annual Rev Psychol. 2008;59(1):255-278.
6. Eva K, Norman GR, Brooks LR. Forward reasoning as a hallmark of expertise in medicine. Adv Psychol Res. 2002;(8):25-40.
7. Davison GC, Vogel RS, Coffman SG. Think-aloud approaches to cognitive assessment and the articulated thoughts in simulated situations paradigm. J Consult Clin Psychol. 1997;65(6):950-958.
8. Ruiter DJ, Kesteren MTR, Fernandez G. How to achieve synergy between medical education and cognitive neuroscience? An exercise on prior knowledge in understanding. Adv Health Sci Educ Theory Pract. 2010;17(2):225-240.
9. Melo M, Scarpin DJ, Amaro E, et al. How doctors generate diagnostic hypotheses: a study of radiological diagnosis with functional magnetic resonance imaging. Aleman A, ed. PLoS ONE. 2011;6(12):e28752.
10. Bahrami P, Graham SJ, Grantcharov TP, et al. Neuroanatomical correlates of laparoscopic surgery training. Surg Endosc. 2014;28(7):2189-2198.
11. Durning SJ, Capaldi VF II, Artino AR Jr., et al. A pilot study exploring the relationship between internists’ self-reported sleepiness, performance on multiple-choice exam items and prefrontal cortex activity. Med Teach. 2014:1-7.
12. Downar J, Bhatt M, Montague PR. Neural Correlates of Effective Learning in Experienced Medical Decision-Makers. Lauwereyns J, ed. PLoS ONE. 2011;6(11):e27768.
13. Durning SJ, Costanzo ME, Artino AR. Neural basis of nonanalytical reasoning expertise during clinical evaluation. Brain and Cognition, 2015.
95
14. Heit E, Rotello CM, Hayes BK. Relations Between Memory and Reasoning. Elsevier Inc.; 2012:57-101.
15. Süß HM, Oberauer K, Wittmann W, Wilhelm O. Working-memory capacity explains reasoning ability—and a little bit more. Intelligence. 2002.
16. Ruff CC, Knauff M, Fangmeier T, Spreer J. Reasoning and working memory: common and distinct neuronal processes. Neuropsychologia. 2003;41(9):1241-1253.
17. Baddeley A. Working memory. Science. 1992;255(5044):556-559.
18. Cowan N. Chapter 20 What are the differences between long-term, short-term, and working memory? In: Essence of Memory.Vol 169. Progress in Brain Research. Elsevier; 2008:323-338.
19. Goel V. Anatomy of deductive reasoning. Trends in Cognitive Sciences. 2007;11(10):435-441.
20. Lee D, Rushworth MFS, Walton ME, Watanabe M, Sakagami M. Functional Specialization of the Primate Frontal Cortex during Decision Making. Journal of Neuroscience. 2007;27(31):8170-8173.
21. Binder JR, Desai RH, Graves WW, Conant LL. Where is the semantic system? A critical review and meta-analysis of 120 functional neuroimaging studies. Cerebral Cortex. 2009;19(12):2767-2796.
22. Squire L, Zola S. Structure and function of declarative and nondeclarative memory systems. Proc Natl Acad Sci U S A. 1996;93(24):13515-13522.
23. Ferstl EC, Neumann J, Bogler C, Cramon von DY. The extended language network: A meta-analysis of neuroimaging studies on text comprehension. Hum Brain Mapp. 2008;29(5):581-593.
24. Rottschy C, Langner R, Dogan I, et al. Modelling neural correlates of working memory: a coordinate-based meta-analysis. Neuroimage. 2012;60(1):830-846.
25. Ernst M, Paulus MP. Neurobiology of Decision Making: A Selective Review from a Neurocognitive and Clinical Perspective. Biological Psychiatry. 2005;58(8):597-
26. Moss J, Schunn CD, Schneider W, McNamara DS, VanLehn K. The neural correlates of strategic reading comprehension: Cognitive control and discourse comprehension. Neuroimage. 2011;58(2):675-686.
27. Goel V. Anatomy of deductive reasoning. Trends in Cognitive Sciences. 2007;11(10):435-441.
28. Cabeza R, Kingstone A. Handbook of Functional Neuroimaging of Cognition. Mit
96
Press; 2006.
29. Croskerry P. A universal model of diagnostic reasoning. Acad Med. 2009;84(8):1022-1028.
30. Durning SJ, Elnicki ME, Gruppen L, Torre D, Hemmer PA. AMEE 2009 spotlight on educational research. Med Teach. 2010;32(4):340-342.
31. Kurzweil R. How to Create a Mind. Penguin; 2012.
32. Baars B, Gage N. Cognition, Brain, and Consciousness. 2nd ed. Boston: Elsevier Ltd; 2010.
33. Binder JR, Desai RH, Graves WW, Conant LL. Where Is the Semantic System? A Critical Review and Meta-Analysis of 120 Functional Neuroimaging Studies. Cerebral Cortex. 2009;19(12):2767-2796.
34. Baddeley A. Exploring the central executive. The Quarterly Journal of Experimental Psychology. 1996;49(1):5-28.
35. Baddeley A, Eysenck MW, Anderson MC. Memory. Psychology Press; 2014.
36. Collette F, Van der Linden M. Brain imaging of the central executive component of working memory. Neurosci Biobehav Rev. 2002;26(2):105-125.
37. Baddeley A. The episodic buffer: a new component of working memory? Trends in Cognitive Sciences. 2000;4(11):417-423.
38. Baddeley AD, Hitch G. Working Memory. In: Vol 8. Psychology of Learning and Motivation. Elsevier; 1974:47-89.
39. Carpenter PA, Just MA, Reichle ED. Working memory and executive function: evidence from neuroimaging. Curr Opin Neurobiol. 2000;10(2):195-199.
40. Sanfey AG, Chang LJ. Multiple Systems in Decision Making. Annals of the New York Academy of Sciences. 2008;1128(1):53-62.
41. Barbey AK. Architecture of explanatory inference in the human prefrontal cortex. Frontiers in Psychology. 2011:1-9.
42. Rosler F, Ranganath C, Röder B, Kluwe R. Neuroimaging of Human Memory. OUP Oxford; 2009.
43. D'Esposito M. From cognitive to neural models of working memory. Philosophical Transactions of the Royal Society B: Biological Sciences. 2007;362(1481):761-772.
44. Baddeley ADHG. Working Memory. G.H. Bower (Ed.). New York: Academic Press;
97
1974:47-89.
45. McRae K, Jones M. Semantic memory. The Oxford handbook of cognitive psychology. 2012.
46. Schumann JH, Crowell SE, Jones NE, Lee N, Schuchert SA. The Neurobiology of Learning. Routledge; 2014.
47. Goel V, Buchel C, Frith C, Dolan RJ. Dissociation of Mechanisms Underlying Syllogistic Reasoning. Neuroimage. 2000;12(5):504-514.
48. Martin A, Chao LL. Semantic memory and the brain: structure and processes. Curr Opin Neurobiol. 2001;11(2):194-201.
49. Moscovitch M, Rosenbaum RS, Gilboa A, et al. Functional neuroanatomy of remote episodic, semantic and spatial memory: a unified account based on multiple trace theory. J Anat. 2005;207(1):35-66.
50. Euston DR, Gruber AJ, McNaughton BL. The Role of Medial Prefrontal Cortex in Memory and Decision Making. Neuron. 2012;76(6):1057-1070.
51. Collette F, Hogge M, Salmon E, Van der Linden M. Exploration of the neural substrates of executive functioning by functional neuroimaging. Neuroscience. 2006;139(1):209-221.
52. Atkinson RC, Shiffrin RM. Human Memory: a Proposed System and Its Control Processes. Stanford University. Institute for Mathematical Studies in the Social Sciences; 1967.
53. Squire LR, Knowlton B, Musen G. The structure and organization of memory. Annu Rev Psychol. 1993;44:453-495.
54. Norman G. Research in clinical reasoning: past history and current trends. Medical Education. 2005;39(4):418-427.
55. Simmons B. Clinical reasoning: concept analysis. J Adv Nurs. 2010;66(5):1151-1158.
56. Balleine BW. The Neural Basis of Choice and Decision Making. Journal of Neuroscience. 2007;27(31):8159-8160.
57. Schmidt HGH, Norman GRG, Boshuizen HPH. A cognitive perspective on medical expertise: theory and implication. Acad Med. 1990;65(10):611-621.
58. McLaughlin K, Coderre S, Mortis G, Fick G, Mandin H. Can Concept Sorting Provide a Reliable, Valid and Sensitive Measure of Medical Knowledge Structure? Adv Health Sci Educ Theory Pract. 2006;12(3):265-278.
98
59. Davison GC, Navarre SG, Vogel RS. The articulated thoughts in simulated situations paradigm: A think-aloud approach to cognitive assessment. Current Directions in Psychological Science. 1995;4(1):29-33.
60. Coderre S, Mandin H, Harasym PH, Fick GH. Diagnostic reasoning strategies and diagnostic success. Medical Education. 2003;37(8):695-703.
61. Boshuizen H. Does Practice Make Perfect? Professional learning: Gaps and transitions on the way from Novice to Expert. Innovation in change in professional education. 2004 (2): 73-95.
62. Charlin B, Tardif J, Boshuizen HP. Scripts and medical diagnostic knowledge: theory and applications for clinical reasoning instruction and research. Academic Medicine. 2000;75(2):182-190.
63. Novak K, Mandin H, Wilcox E, McLaughlin K. Using a conceptual framework during learning attenuates the loss of expert-type knowledge structure. BMC Med Educ. 2006;6:37.
64. Coderre S, Jenkins D, McLaughlin K. Qualitative differences in knowledge structure are associated with diagnostic performance in medical students. Adv Health Sci Educ Theory Pract. 2009;14(5):677-684.
65. Gruppen LD, Frohna AZ. Clinical Reasoning. In: International Handbook of Research in Medical Education. Dordrecht: Springer Netherlands; 2002:205-230.
66. Custers EJ, Stuyt PM, De Vries Robbé PF. Clinical problem analysis (CPA): a systematic approach to teaching complex medical problem solving. Academic Medicine. 2000;75(3):291-297.
67. Kulatunga-Moruzi C, Brooks LR, Norman GR. Coordination of Analytic and Similarity-Based Processing Strategies and Expertise in Dermatological Diagnosis. Teaching and Learning in Medicine. 2001;13(2):110-116.
68. Ericsson KA, Charness N, Feltovich PJ, Hoffman RR. The Cambridge Handbook of Expertise and Expert Performance. Cambridge University Press; 2006.
69. Patel VL, Evans DA, Groen GJ. Biomedical knowledge and clinical reasoning. Cognitive science in medicine. The MIT Press, 1989.
70. Raman M, McLaughlin K, Violato C, Rostom A, Allard JP, Coderre S. Teaching in small portions dispersed over time enhances long-term knowledge retention. Med Teach. 2010;32(3):250-255.
71. Larsen DP, Butler AC, Roediger HL III. Comparative effects of test-enhanced learning and self-explanation on long-term retention. Medical Education. 2013;47(7):674-682.
99
72. Elieson SW, Papa FJ. The effects of various knowledge formats on diagnostic performance. Academic Medicine. 1994;69(10):S81-S83.
73. Ericsson KA, Charness N, Feltovich PJ, Hoffman RR. The Cambridge Handbook of Expertise and Expert Performance. Cambridge University Press; 2006.
74. Coderre S, Wright B, McLaughlin K. To think is good: querying an initial hypothesis reduces diagnostic error in medical students. Acad Med. 2010;85(7):1125-1129.
75. Pelaccia T, Tardif J, Triby E, Charlin B. An analysis of clinical reasoning through a recent and comprehensive approach: the dual-process theory. Medical Education Online. 2011;16(0).
76. Norman GR, Brooks LR, Colle CL, Hatala RM. The Benefit of Diagnostic Hypotheses in Clinical Reasoning: Experimental Study of an Instructional Intervention for Forward and Backward Reasoning. Cognition and Instruction. 1999;17(4):433-448.
77. Hobus PP, Schmidt HG, Boshuizen HP, Patel VL. Contextual factors in the activation of first diagnostic hypotheses: expert-novice differences. Medical Education. 1987;21(6):471-476.
78. Norman GR, Brooks LR. The non-analytical basis of clinical reasoning. Adv Health Sci Educ Theory Pract. 1997;2(2):173-184.
79. Klein, G. A Recognition-Primed Decision (RPD) Model of Rapid Decision Making. U.S. Army Research Institute for the Behavioral and Social Sciences, 1993.
80. Eva KW, Brooks LR. The Under-weighting of Implicitly Generated Diagnoses. Academic Medicine. 2000;75(Supplement):S81-S83.
81. Eva KW. What every teacher needs to know about clinical reasoning. Medical Education. 2005;39(1):98-106.
82. Kulatunga-Moruzi C, Brooks LR, Norman GR. Coordination of Analytic and Similarity-Based Processing Strategies and Expertise in Dermatological Diagnosis. Teaching and Learning in Medicine. 2001;13(2):110-116.
83. Evans J, Stanovich K. Dual-Process Theories of Higher Cognition: Advancing the Debate. Perspectives on Psychological Science. 2013;8(3):223-241.
84. Croskerry P. A universal model of diagnostic reasoning. Acad Med. 2009;84(8):1022-1028.
85. Haller S, Radue EW. What Is Different about a Radiologist's Brain? Radiology. 2005;236(3):983-989.
100
86. Durning SJ. Functional neuroimaging correlates of burnout among internal medicine residents and faculty members. 2013:1-7.
87. Savoy RL. History and future directions of human brain mapping and functional neuroimaging. Acta Psychol (Amst). 2001;107(1-3):9-42.
88. Heeger DJ, Ress D. What does fMRI tell us about neuronal activity? Nature. 2002;3(2):142-151.
89. Amaro E Jr., Barker GJ. Study design in fMRI: Basic principles. Brain and Cognition. 2006;60(3):220-232.
90. Huettel SA, Song AW, McCarthy G. Functional Magnetic Resonance Imaging. Sinauer Associates Incorporated; 2009.
91. Poldrack RA, Mumford JA, Nichols TE. Handbook of Functional MRI Data Analysis. Cambridge University Press; 2011.
92. Ashby FG. Statistical Analysis of FMRI Data. MIT Press; 2011.
93. Derrfuss J, Mar RA. Lost in localization: The need for a universal coordinate database. Neuroimage. 2009;48(1):1-7.
94. Logothetis NK. The underpinnings of the BOLD functional magnetic resonance imaging signal. J Neurosci. 2003;23(10):3963-3971.
95. Attwell D, Iadecola C. The neural basis of functional brain imaging signals. Trends Neurosci. 2002;25(12):621-625.
96. Raichle ME, Mintun MA. Brain work and brain imaging. Annu Rev Neurosci. 2006;29(1):449-476.
97. Aguirre GK, D'Esposito M. 30 Experimental designs for brain fMRI. BOLD. 2000:369-379.
98. Guo Q, Hall, McKinnon, Thabane L, Goeree R, Pullenayegum. Setting sample size using cost efficiency in fMRI studies. OAMS. 2012:33-39.
99. Desmond JE, Glover GH. Estimating sample size in functional MRI (fMRI) neuroimaging studies: statistical power analyses. Journal of Neuroscience Methods. 2002;118(2):115-128.
100. Friston KJ, Holmes AP, Worsley KJ. How many subjects constitute a study? Neuroimage. 1999;10(1):1-5.
101. Beckmann CF, Jenkinson M, Smith SM. General multilevel linear modeling for group analysis in FMRI. Neuroimage. 2003;20(2):1052-1063.
101
102. Smith SM, Jenkinson M, Woolrich MW, et al. Advances in functional and structural MR image analysis and implementation as FSL. Neuroimage. 2004;23:S208-S219.
103. Poldrack RA. Region of interest analysis for fMRI. Soc Cogn Affect Neurosci. 2007;2(1):67-70.
104. Savoy RL, Ravicz ME, Gollub R. Functional MRI. In: Berlin: Springer; 1999:347-365.
105. Ericsson KA, Kintsch W. Long-term working memory. Psychological Review. 1995;102(2):211-245.
106. Ranganath C, Johnson M, D'Esposito M. Prefrontal activity associated with working memory and episodic long-term memory. Neuropsychologia. 2003;41(3):378-389.
107. Simons JS, Spiers HJ. Prefrontal and medial temporal lobe interactions in long-term memory. Nature Review Neuroscience. 2003;4(8):637-648.
108. Oldfield R. The assessment and analysis of handedness: the Edinburgh inventory. Neuropsychologia. 1971;9(1):97-113.
109. Savoy RL. Using small numbers of subjects in fMRI-based research. IEEE Eng Med Biol Mag. 2006;25(2):52-59.
110. Jenkinson M. Fast, automated, N-dimensional phase-unwrapping algorithm. Magn Reson Med. 2003;49(1):193-197.
111. Jenkinson M, Smith S. A global optimisation method for robust affine registration of brain images. Med Image Anal. 2001;5(2):143-156.
112. Worsley KJ, Evans AC, Marrett S. A three-dimensional statistical analysis for CBF activation studies in human brain. Journal of Cerebral Blood Flow & Metabolism. 1992.
113. Velanova K, Jacoby LL, Wheeler ME, McAvoy MP, Petersen SE, Buckner RL. Functional-anatomic correlates of sustained and transient processing components engaged during controlled retrieval. Journal of Neuroscience. 2003;23(24):8460-8470.
114. Guida A, Gobet F, Tardieu H, Nicolas S. How chunks, long-term working memory and templates offer a cognitive explanation for neuroimaging data on expertise acquisition: A two-stage framework. Brain and Cognition. 2012;79(3):221-244.
115. Rogers T, Hocking J, Noppeney U, et al. Anterior temporal cortex and semantic memory: Reconciling findings from neuropsychology and functional imaging. Cognitive Affect Behavioural Neuroscience. 2006;6(3):201-213.
102
116. Rogers T, Lambon Ralph MA, Garrard P, et al. Structure and deterioration of semantic memory: a neuropsychological and computational investigation. Psychological Review. 2004;111(1):205-235.
117. Croskerry P. The theory and practice of clinical decision-making. Can J Anesth/J Can Anesth. 2005;52:R1-R8.
118. Higgs J, Jones M, Loftus S, Christensen N. Clincal Reasoning in the Health Professions. 3rd ed. Boston: Elsevier; 2008.
119. Ross KG, Klein GA, Thunholm P, Schmitt JF, Baxter HC. The recognition-primed decision model. Military Review; 2004.
120. Eva KW. What every teacher needs to know about clinical reasoning. Medical Education. 2005;39(1):98-106.
121. Evans JSBT. In two minds: dual-process accounts of reasoning. Trends in Cognitive Sciences. 2003;7(10):454-459.
122. Durning SJ, Costanzo M, Artino AR, et al. Using functional magnetic resonance imaging to improve how we understand, teach, and assess clinical reasoning. J Contin Educ Health Prof. 2014;34(1):76-82.
123. Melo M, Scarpin DJ, Amaro E, et al. How Doctors Generate Diagnostic Hypotheses: A Study of Radiological Diagnosis with Functional Magnetic Resonance Imaging. Aleman A, ed. PLoS ONE. 2011;6(12):e28752.
124. Haller S, Radue EW. What Is Different about a Radiologist's Brain? Radiology. 2005;236(3):983-989.
125. Bahrami P, Graham SJ, Grantcharov TP, et al. Neuroanatomical correlates of laparoscopic surgery training. Surg Endosc. 2014;28(7):2189-2198.
126. Krawczyk DC. Contributions of the prefrontal cortex to the neural basis of human decision making. Neurosci Biobehav Rev. 2002;26(6):631-664.
127. Jenkinson M, Smith S. A global optimisation method for robust affine registration of brain images. Med Image Anal. 2001;5(2):143-156.
128. Worsley KJ, Marrett S, Neelin P, Vandal AC, Friston KJ, Evans AC. A unified statistical approach for determining significant signals in images of cerebral activation. Human Brain Mapping. 1996;4(1):58-73.
129. Fangmeier T, Knauff M, Ruff CC, Sloutsky V. FMRI evidence for a three-stage model of deductive reasoning. J Cogn Neurosci. 2006;18(3):320-334.
130. McElroy T, Seta JJ. On the other hand am I rational? Hemispheric activation and the
103
framing effect. Brain and Cognition. 2004;55(3):572-580.
131. Wolford G, Miller MB, Gazzaniga M. The left hemisphere's role in hypothesis formation. The Journal of Neuroscience. 2000.
132. Croskerry P. Clinical cognition and diagnostic error: applications of a dual process model of reasoning. Advanced Health Science Education Theory Practice. 2009;14(S1):27-35.
133. Gregory A, Walker I, McLaughlin K, Peets AD. Both preparing to teach and teaching positively impact learning outcomes for peer teachers. Med Teach. 2011;33(8):e417-e422.
134. Logothetis NK. What we can do and what we cannot do with fMRI. Nature. 2008;453(7197):869-878.
104
Appendix A: MRI SCREENING FORM
105
Appendix B: PRE SCREENING QUESTIONNAIRE
Please indicate all educational achievements that apply:
o Baccalaureate Degree ☐ Masters Degree ☐Doctor of Philosophy
o Medical Degree ☐ Advanced Specialization in Medicine
Please specify the areas of medicine you specialize in:
________________________________________________________________________________
Overall years of clinical experience (including residency):
☐0 years ☐11-20 years ☐31-40 years
☐1-10 years ☐21-30 years ☐41-50 years
Years of practice in your area of clinical specialization:
☐0 years or N/A ☐11-20 years ☐31-40 years
☐1-10 years ☐21-30 years ☐41-50 years
Age: _____ Sex: ☐Male ☐Female Dominant Hand: ☐Right ☐Left
How many hours of sleep have you had in the past 24hrs? ___________________
Does your clinical position require you to formally teach?
Does your clinical position require you to informally teach?
Please indicate by checkmark if you have a history of any of the following:
o Previous diagnosis of traumatic brain injury
o Non medically induced loss of consciousness for >10 minutes
o Diagnosis of psychiatric illness requiring medical treatment
o Previous or current use of psychotropic drugs
o History of epilepsy or seizures
o History of cerebral stroke
o History of cerebral aneurysm
o Current use of calcium channel blockers
o History of claustrophobia
o Any other diagnosis of neurovascular or neurologic illness/injury
106
Appendix C: POST SCREENING QUESTIONNAIRE
We would like to know how your participation in our study went. If you could please take a moment to answer the following questions and provide any feedback, it would be greatly appreciated. 1. Were the instructions provided to you for your participation clear and understandable?
o Yes o No If no, please let us know what was unclear?
2. After having been through the scanning sequences, were the instructions accurate and reflective of what you experienced?
o Yes o No If no, please let us know what we could have better prepared you for?
3. Could you hear instructions clearly when in the MR scanner?
o Yes o No Any further comments:
4. Did you feel like you could clearly communicate to the research coordinator and/or
technicians if you needed to? o Yes o No Any further comments:
5. Did you feel comfortable during your scan? o Yes o No Any further comments:
6. Did you have enough time to read each clinical case? o Yes o No Any further comments:
7. Did you have enough time to answer the questions after each clinical case? o Yes o No Any further comments:
107
8. When answering questions, did you feel it was easy to use the hand boxes to submit your answers?
o Yes o No Any further comments:
9. Do you have any other feedback you would like to provide?