Ga¨ el Varoquaux Manuscript submitted A. Abraham
Jul 29, 2015
Inter-site autism biomarkersfrom resting-state fMRI
Gael Varoquaux
Manuscript submitted A. Abraham
Predictive biomarkers from restSuitable for diminished patients
Probing intrinsic brain structurePredicting pathologies as a validation proxy
Multi-site large autism dataset: ABIDEAutism Spectrum Disorder [Di Martino... 2014]
⇒ Patient/Control classification
16 sites
∼ 1200 subjects (900 after QC)
G Varoquaux 2
Experiments: autism predictionSeen sites: predicting to new subjects
Training set Testing set
Unseen sites: predicting to new sites
Training set Testing set
Training set Test setG Varoquaux 3
A connectome classification pipeline
RS-fMRI
Functionalconnectivity
Time series
24
3
1
Diagnosis
ROIs
1 ROI definition2 Time-series extraction3 Connectivity matrices4 Supervised learning
Prediction accuracy (%)Seen sites 67±3
Unseen sites 67±5
What is important to predict?
G Varoquaux 4
A connectome classification pipeline
RS-fMRI
Functionalconnectivity
Time series
24
3
1
Diagnosis
ROIs
1 ROI definition2 Time-series extraction3 Connectivity matrices4 Supervised learning
Prediction accuracy (%)Seen sites 67±3
Unseen sites 67±5
What is important to predict?
G Varoquaux 4
A connectome classification pipeline
RS-fMRI
Functionalconnectivity
Time series
24
3
1
Diagnosis
ROIs
1 ROI definition2 Time-series extraction3 Connectivity matrices4 Supervised learning
Prediction accuracy (%)Seen sites 67±3
Unseen sites 67±5
What is important to predict?
G Varoquaux 4
1 ROI definition
RS-fMRI
Functionalconnectivity
Time series
24
3
1
Diagnosis
ROIs
1 ROI definition2 Time-series extraction3 Connectivity matrices4 Supervised learning
G Varoquaux 5
1 ROI definition: Learning functional regions
Harvard Oxford
Anatomical atlases do not resolve functionalstructures
Where is the default mode network?
G Varoquaux 6
1 ROI definition: Clustering approaches
K-MeansRelated to [Yeo... 2011]
Normalized cuts[Craddock... 2012]
Ward clustering[Thirion... 2014]
G Varoquaux 7
1 ROI definition: Linear decomposition approaches
Model: Observing linear mixtures of networks at rest
Time courses
G Varoquaux 8
1 ROI definition: Linear decomposition approaches
Model: Observing linear mixtures of networks at rest
Time courses
Language
G Varoquaux 8
1 ROI definition: Linear decomposition approaches
Model: Observing linear mixtures of networks at rest
Time courses
Audio
G Varoquaux 8
1 ROI definition: Linear decomposition approaches
Model: Observing linear mixtures of networks at rest
Time courses
Visual
G Varoquaux 8
1 ROI definition: Linear decomposition approaches
Model: Observing linear mixtures of networks at rest
Time courses
Dorsal Att.
G Varoquaux 8
1 ROI definition: Linear decomposition approaches
Model: Observing linear mixtures of networks at rest
Time courses
Motor
G Varoquaux 8
1 ROI definition: Linear decomposition approaches
Model: Observing linear mixtures of networks at rest
Time courses
Salience
G Varoquaux 8
1 ROI definition: Linear decomposition approaches
Model: Observing linear mixtures of networks at rest
Time courses
Observing a mixtureHow to unmix networks?
ICA: minimize mutual information across SSparse decompositions: sparse penalty on maps
G Varoquaux 8
1 ROI definition: MSDL linear decomposition
[Varoquaux... 2011, Abraham... 2013]
MSDL: multi-subject dictionary learning
Sparse
Multi-subject
Subject maps + Population-levelatlas
Spatial penalty(total variation)
L R
y=-23 x=0
L R
z=18
L R
y=-78 x=2
L R
z=4
Ventricular system Visual CortexL R
y=-47 x=51
L R
z=11
L R
y=-17 x=-47
L R
z=8
TPJ Auditory Network
G Varoquaux 9
1 ROI definition: MSDL linear decomposition
[Varoquaux... 2011, Abraham... 2013]
MSDL: multi-subject dictionary learning
Sparse
Multi-subject
Subject maps + Population-levelatlas
Spatial penalty(total variation)
L R
y=-23 x=0
L R
z=18
L R
y=-78 x=2
L R
z=4
Ventricular system Visual CortexL R
y=-47 x=51
L R
z=11
L R
y=-17 x=-47
L R
z=8
TPJ Auditory Network
G Varoquaux 9
1 ROI definition: different optionsAt
lase
s
Craddock2011 Ncuts
Harvard-Oxford
Yeo 2011
Regi
onle
arni
ng
K-Means Ward MSDL ICA
G Varoquaux 10
1 ROI definition: impact of choice
RS-fMRI
Functionalconnectivity
Time series
24
3
1
Diagnosis
ROIs
1 ROI definition2 Time-series extraction3 Connectivity matrices4 Supervised learning
G Varoquaux 11
2 Time-series extraction
Time series
2
RS-fMRI
Functionalconnectivity
43
1
Diagnosis
ROIs
1 ROI definition2 Time-series extraction3 Connectivity matrices4 Supervised learning
Remove motion regressorsCompcorrGlobal mean regression?Empirically: different ways work
G Varoquaux 12
2 Time-series extraction
Time series
2
RS-fMRI
Functionalconnectivity
43
1
Diagnosis
ROIs
1 ROI definition2 Time-series extraction3 Connectivity matrices4 Supervised learning
Remove motion regressorsCompcorrGlobal mean regression?Empirically: different ways work
G Varoquaux 12
3 Functional-connectivity matrix
Time series
2
RS-fMRI
41
Diagnosis
ROIs Functionalconnectivity
3
1 ROI definition2 Time-series extraction3 Connectivity matrices4 Supervised learning
Correlation matrix?Partial correlation matrix?Tangent-space embedding?
[Varoquaux... 2010]
G Varoquaux 13
3 Functional-connectivity matrix
Time series
2
RS-fMRI
41
Diagnosis
ROIs Functionalconnectivity
3
1 ROI definition2 Time-series extraction3 Connectivity matrices4 Supervised learning
Correlation matrix?Partial correlation matrix?Tangent-space embedding?
[Varoquaux... 2010]
G Varoquaux 13
3 Functional-connectivity matrix: different options
0 5 10 15 20 25
0
5
10
15
20
25 Control0 5 10 15 20 25
0
5
10
15
20
25 Control0 5 10 15 20 25
0
5
10
15
20
25 Control0 5 10 15 20 25
0
5
10
15
20
25Large lesion
Correlation matrices
0 5 10 15 20 25
0
5
10
15
20
25 Control0 5 10 15 20 25
0
5
10
15
20
25 Control0 5 10 15 20 25
0
5
10
15
20
25 Control0 5 10 15 20 25
0
5
10
15
20
25Large lesion
Partial correlation matrices
0 5 10 15 20 25
0
5
10
15
20
25 Control0 5 10 15 20 25
0
5
10
15
20
25 Control0 5 10 15 20 25
0
5
10
15
20
25 Control0 5 10 15 20 25
0
5
10
15
20
25Large lesion
Tangent-space embedding[varoquaux 2010]
G Varoquaux 14
3 Functional-connectivity matrix: impact of choice
Time series
2
RS-fMRI
41
Diagnosis
ROIs Functionalconnectivity
3
1 ROI definition2 Time-series extraction3 Connectivity matrices4 Supervised learning
G Varoquaux 15
4 Supervised learning method
Functionalconnectivity
Time series
34
Diagnosis
2
RS-fMRI
1 ROIs
1 ROI definition2 Time-series extraction3 Connectivity matrices4 Supervised learning
Standard supervised learningRidge classifierSVC `2 penalizedSVC `1 penalized
G Varoquaux 16
4 Supervised learning method
Functionalconnectivity
Time series
34
Diagnosis
2
RS-fMRI
1 ROIs
1 ROI definition2 Time-series extraction3 Connectivity matrices4 Supervised learning
Standard supervised learningRidge classifierSVC `2 penalizedSVC `1 penalized
G Varoquaux 16
4 Supervised learning method: impact of choice
Functionalconnectivity
Time series
34
Diagnosis
2
RS-fMRI
1 ROIs
1 ROI definition2 Time-series extraction3 Connectivity matrices4 Supervised learning
G Varoquaux 17
Importance of pipeline steps
RS-fMRI
Functionalconnectivity
Time series
24
3
1
Diagnosis
ROIs
1 ROI definition2 Time-series extraction3 Connectivity matrices4 Supervised learning
G Varoquaux 18
Importance of pipeline steps
RS-fMRI
Functionalconnectivity
Time series
24
3
1
Diagnosis
ROIs
1 ROI definition2 Time-series extraction3 Connectivity matrices4 Supervised learning
G Varoquaux 18
How general are these conclusions?Different subsets:
Set Card. Sites Selection criteriaA 871 17 All subjects after QCB 736 11 Remove 6 smallest sitesC 639 14 A without left handed and womenD 420 14 C between 9 and 18 years oldE 226 3 D from 3 biggest sites
Exploring variabilityFrom most heterogeneousto most homogeneous
G Varoquaux 19
Across different sets of subjects
RS-fMRI
Functionalconnectivity
Time series
24
3
1
Diagnosis
ROIs
1 ROI definition2 Time-series extraction3 Connectivity matrices4 Supervised learning
MSDL outperform alternatives
Intra-siteInter-site
G Varoquaux 20
Across different sets of subjects
RS-fMRI
Functionalconnectivity
Time series
24
3
1
Diagnosis
ROIs
1 ROI definition2 Time-series extraction3 Connectivity matrices4 Supervised learning
Tangent space outperform alternatives
Intra-siteInter-site
G Varoquaux 20
Across different sets of subjects
RS-fMRI
Functionalconnectivity
Time series
24
3
1
Diagnosis
ROIs
1 ROI definition2 Time-series extraction3 Connectivity matrices4 Supervised learning
`2 classifiers outperform alternatives
Intra-siteInter-site
G Varoquaux 20
MSDL atlas
L R
y=-54 x=0
L R
z=3
L RL R
L RL R
L RL R
More data is better
Multivariate processing of a 1Tb of heterogeneous datais worth the trouble
@GaelVaroquaux
Intersite biomarkers from rest [Abraham submitted]
Prediction of autismCarries out across sites
Recipe for a good pipelineChoice of regions critical (MSDL to learn them)Tangent-space embeddingStandard SVM
Definition of regions MSDLLinear decomposition + sparsity
+ total-variation
ni
@GaelVaroquaux
Intersite biomarkers from rest [Abraham submitted]
Prediction of autismCarries out across sites
Recipe for a good pipelineChoice of regions critical (MSDL to learn them)Tangent-space embeddingStandard SVMConnectome structure not central for prediction
Definition of regions MSDLLinear decomposition + sparsity
+ total-variation
ni
@GaelVaroquaux
Intersite biomarkers from rest [Abraham submitted]
Prediction of autismCarries out across sites
Recipe for a good pipelineChoice of regions critical (MSDL to learn them)Tangent-space embeddingStandard SVM
Definition of regions MSDLLinear decomposition + sparsity
+ total-variation
ni
References I
A. Abraham, E. Dohmatob, B. Thirion, D. Samaras, andG. Varoquaux. Extracting brain regions from rest fMRI withtotal-variation constrained dictionary learning. In MICCAI, page607. 2013.
R. C. Craddock, G. A. James, P. E. Holtzheimer, X. P. Hu, andH. S. Mayberg. A whole brain fMRI atlas generated via spatiallyconstrained spectral clustering. Human brain mapping, 33(8):1914–1928, 2012.
A. Di Martino, C.-G. Yan, Q. Li, E. Denio, F. X. Castellanos,K. Alaerts, J. S. Anderson, M. Assaf, S. Y. Bookheimer,M. Dapretto, ... The autism brain imaging data exchange:towards a large-scale evaluation of the intrinsic brainarchitecture in autism. Molecular psychiatry, 19:659, 2014.
B. Thirion, G. Varoquaux, E. Dohmatob, and J. Poline. WhichfMRI clustering gives good brain parcellations? Name: Frontiersin Neuroscience, 8:167, 2014.
References IIG. Varoquaux, F. Baronnet, A. Kleinschmidt, P. Fillard, and
B. Thirion. Detection of brain functional-connectivity differencein post-stroke patients using group-level covariance modeling. InMICCAI, pages 200–208. 2010.
G. Varoquaux, A. Gramfort, F. Pedregosa, V. Michel, andB. Thirion. Multi-subject dictionary learning to segment an atlasof brain spontaneous activity. In Inf Proc Med Imag, pages562–573, 2011.
B. Yeo, F. Krienen, J. Sepulcre, M. Sabuncu, ... The organizationof the human cerebral cortex estimated by intrinsic functionalconnectivity. J Neurophysio, 106:1125, 2011.