discovering students’ viewing behavior during a practical...
TRANSCRIPT
Discovering students’ viewing behavior during a practical exam in oral pathology using software-based view path tracking for whole slide images SLAWOMIR WALKOWSKI1, JOHAN LUNDIN2, JANUSZ SZYMAS3 , MIKAEL LUNDIN2
1Poznan University of Technology, Poland 2Institute for Molecular Medicine Finland FIMM, 3Department of Clinical Pathology, Poznan University of Medical Sciences, Poland
Background
Whole slide images (WSIs) may not only replace the way of viewing histological slides but also provide new insights into how the slides are viewed
Especially, WSIs may help uncover viewing patterns among students who take a practical exam in oral pathology
Goals
Create software infrastructure for a view path tracking method for whole slide images (WSIs), which is scalable to large number of students simultanuously taking an exam and does not require specialized equipment
Collect data during the exam, then discover and analyze students’ viewing behavior while interpreting WSIs
View path tracking Each time a student stopped panning and zooming and the whole view
field image data was loaded, a record was sent to the database ~74,000 records sent in total during an exam Average of ~750 view fields per student in an exam session Average of ~15 view fields per WSI per student in an exam session
Database Images
100% loaded
View field record
• coordinates viewed • timestamp • student and question ID
Server
Visualization – view path comparison
Verrucous Carcinoma Answered correctly
Answered incorrectly
All answers
Measures A set of 26 statistics was designed and calculated to numerically compare
and describe viewing behavior of each student in each question
1. Total viewing time 2. Number of view steps
Measures
AverageZoom 2.25 Dispersion 0.26 MaxZoom 3.85 MinZoom 0.35 RoiViewed 0.01 … …
View fields per student per question
The statistics include:
Measures 3. ROI viewed
0.012 < 0.624
Cheilitis actinica
Weighted by view field viewing time
First (default) view field skipped
Measures
2.517 < 17.926
Pseudoepitheliomatous Hyperplasia 4. Average zoom level Weighted by view field viewing time
Benign Reactive Keratosis One student not focused on ROI
Viewing Time # View Fields ROIs Viewed Avg Zoom Dispersion40 27 0.908 21.663 0.11616 14 0.066 20.871 0.061
Follicular Cyst One student uses too low magnification
Viewing Time # View Steps ROI Viewed Avg Zoom Dispersion49 39 0.942 26.635 0.13512 8 0.296 1.972 0.052
46140
2356
0.8040.704
13.97512.071
0.1970.283
Squamous Cell Carcinoma
Average
Dist
ribut
ion
Analyzing answers from all students based on statistics comparison
0% 10% 20% 30% 40% 50% 60% 70%
Viewing Time
0%
20%
40%
60%
# View Fields
0% 10% 20% 30% 40% 50% 60% 70%
ROIs Viewed
-10%
10%
30%
50%
70% Average Zoom
0% 10% 20% 30% 40% 50% 60% 70%
Dispersion
Analyzing multiple questions 26 questions (including 8 with ROIs specified), which had at least 5 correct
and 5 incorrect answers, were compared numerically For each measure, questions with average value higher for correct answers
and questions with average value higher for incorrect answers were counted
0 5 10 15 20 25
Viewing Time
# View Fields
ROI Viewed
Average Zoom
Dispersion
Number of questions
Average measure higher for correct answers
Average measure higher for incorrect answers
Conclusions
View path tracking appeared to be a useful method of discovering viewing behavior of a large group of users, interpreting many WSIs
Gathered data was visualized in a variety of ways, providing many useful insights on both individual viewing behaviors and patterns occurring across multiple students
Automatically calculated statistics reflected some viewing characteristics, confirmed visual observations and allowed generalization of some findings across multiple students and questions