emily finn - brainhack nyc
TRANSCRIPT
Individual differences in functional brain connectivity
Emily S. FinnBrainhack NYCChild Mind Institute2 March 2017
MotivationCan we pick someone out of a crowd based on their functional connectivity profile?
2
3
?1
Scan 1 Scan 2
?
?
Functional connectivity analysisBO
LD
Time (s)
1Shen et al., NeuroImage (2013)
268-node parcellation1
1 2681
268
…
Fisher z score
…
Human Connectome Project• 126 healthy subjects (50 sets of twins)• Age 22-35 years old
MotorMt
EmotionEm
LanguageLg
Day 1
Day 2
Working memoryWM
RestingR1
+
RestingR2
+
Human Connectome Project• 126 healthy subjects (50 sets of twins)• Age 22-35 years old
Motor
EmotionLanguage
Day 1
Day 2
Working memoryResting
Resting
Identification experiments
Subj 1 Subj 2 Subj 3 Subj n
…Database: Rest, day 1
Target: Rest, day 2
r1 r2 r3 rn
Identification experiments
Subj 1 Subj 2 Subj 3 Subj n
…Database: Rest, day 1
Target: Rest, day 2
r1 r2 r3 rn
Identification results
0.93 0.84 0.63
0.72 0.79 0.67
0.64 0.60 0.54
0.94 0.79 0.84
0.57 0.87 0.75
0.79 0.80 0.88
Rest 1
Working memory
Motor
Rest 2 Language Emotion
Chance: ~0.008
Data
base
Target
Rest 2 Language Emotion
Rest 1
Working memory
Motor
Database
Target
ID rate0.5 1.0
Finn, Shen et al., Nat Neurosci (2015)
Identification results
0.93 0.84 0.63
0.72 0.79 0.67
0.64 0.60 0.54
0.94 0.79 0.84
0.57 0.87 0.75
0.79 0.80 0.88
Rest 1
Working memory
Motor
Rest 2 Language Emotion
Chance: ~0.008
Data
base
Target
Rest 2 Language Emotion
Rest 1
Working memory
Motor
Database
Target
ID rate0.5 1.0
Finn, Shen et al., Nat Neurosci (2015)
Why is this important?‣ subject variance > state variance:
the same brain doing two different things looks more similar than two different brains doing the same thing
‣ could relate to behavioral phenotypes
Predicting fluid intelligence• ability to discern patterns• independent of acquired knowledge
Finn, Shen et al., Nat Neurosci (2015)
Ground truthr = 0.50
r = 0.13 r = 0.71
r = 0.81
Less identifiable
More identifiable
More similarLess similar
r = 0.01
“Caricature” “Spotlight”
Between-subject similarity
MotorMot
EmotionEmo
LanguageLan
Working memoryWM
RestR1
+
GamblingGam
RelationalRel
RestR2
+
Day 1
Day 2
SocialSoc
HCP: n = 716, 4:12 per condition
Between-subject similarityNodes
Subjects
Edges
SubjectsWorking memoryWM
EmotionEmo
RestR1
+
1 2
Subjects3
Between-subject similarity
Finn et al., NeuroImage, in pressPart of the special issue on Functional Architecture
Between-subject similarity
Behavioral performance:
Finn et al., NeuroImage, in pressPart of the special issue on Functional Architecture
Between-subject similarity
Target
Data
base
Day 1 Day 2
ID ra
te
Day
1D
ay 2
Replicating identification experiments Conditions that make subjects look more similar to one another actually make better databases for identification
Finn et al., NeuroImage, in pressPart of the special issue on Functional Architecture
Chance ~ 0.001
Between-subject similarity
Ground truth
Less identifiable
More identifiable
More similarLess similar
“Caricature” “Spotlight”
Conditions that make subjects look more similar to one another actually make better databases for identification
Naturalistic tasks
Session 1
Rest Inscapes Ocean’s 11
Session 2
• n = 34 subjects• mean age 24.4 ± 5.1 years
Naturalistic tasks
Session 1
Rest Inscapes Ocean’s 11
Session 2
Intra- versus inter-subject similarity:
Vanderwal et al., in resubmissionAvailable on biorxiv
Naturalistic tasksTimecourse-based Connectivity-based
Session 1 Session 2 Session 1 Session 2
Vanderwal et al., in resubmissionAvailable on biorxiv
Naturalistic tasksHealthy Brain Network Serial Scanning Initiative:• 13 subjects, 12 scans, 4 conditions (rest, 2 movies, Flanker)
Summary
• Functional connectivity profiles:‣ are reliable within individuals‣ are unique across individuals‣ relate to behavior
• Rest = not the optimal condition?‣ Tasks may improve ratio of within- to
between-subject variability‣ Naturalistic tasks are especially intriguing
…
?
Gambling
Emotion
Relational