Machine learning and domain adaptation for cortical thickness in autism

Stand-By Time

Wednesday, June 28, 2017: 12:45 PM  - 2:45 PM 

Submission No:


Submission Type:

Abstract Submission 

On Display:

Wednesday, June 28 & Thursday, June 29 


Jussi Tohka1, Elaheh Moradi2, Budhachandra Khundrakpam3, John Lewis3, Alan Evans3


1University of Eastern Finland, Kuopio, Finland, 2University of Tampere, Tampere, Finland, 3Montreal Neurological Institute, Montreal, Canada

First Author:

Jussi Tohka    -  Lecture Information | Contact Me
University of Eastern Finland
Kuopio, Finland


Machine learning approaches have been widely used for the identification of neuropathology from neuroimaging data. However, these approaches require large samples and suffer from the challenges associated with multi-site, multi-protocol data. We propose a novel approach to address these challenges, and demonstrate its usefulness with the Autism Brain Imaging Data Exchange (ABIDE) database. With this data, we predict symptom severity based on cortical thickness measurements from 156 individuals with autism spectrum disorder (ASD) from four different sites.


The data used in this study were from the ABIDE dataset (Di Martino 2014). The T1-weighted volumes were processed into cortical thickness measures with CIVET, a fully automated structural image analysis pipeline. Cortical thickness (CT) was measured in native space using the linked distance between the two surfaces at 81,924 vertices. After the quality control and excluding ASD subjects with missing ADOS total and module information, we included subjects from sites containing at least 20 subjects. The remaining 156 subjects from 4 different sites were used for predicting severity scores.

The proposed approach consists of two main stages (Moradi 2016, see Figure 1): a domain adaptation stage using partial least squares (PLS) regression to maximize the consistency of imaging data across sites; and a learning stage combining support vector regression (SVR) for regional prediction of severity with elastic-net penalized linear regression for integrating regional predictions into a whole-brain severity prediction. In more detail, we divide the cortical thickness measures into 78 regional subsets based on the AAL atlas and run PLS regression based domain adaptation separately for each region. The regional PLS components are then the input for the SVR with a radial basis function kernel learning 78 region-wise predictors for symptom severity. Finally, the region-wise predictors are combined into a whole brain predictor by elastic-net penalized linear regression. The evaluation was based on multiply nested 10-fold cross validation, which was iterated 100 times.
Supporting Image: graphicalabstract1.png
   ·Figure 1. Overview of the method.


The average cross-validated correlation R between the estimated and observed severity scores among 100 distinct 10-fold CV iterations was 0.51 (standard deviation 0.04, range from 0.39 to 0.63), the average mean absolute error (MAE) was 1.36 (standard deviation 0.05, range from 1.25 to 1.51) and the average coefficient of determination Q2 was 0.26 (standard deviation 0.045, range from 0.13 to 0.39). These values indicated that the proposed approach was able to provide information about the severity of the disease based on structural information of the brain in ASD patients. Fig. 2 shows the scatter plot.

For evaluation of the effectiveness of each stage of the proposed approach, we performed experiments by excluding each stage of the method separately and comparing the accuracy of the predictions obtained this way to the accuracy of the predictions of the complete method. When the PLS-based domain adaptation was substituted by principal component analysis (PCA), the average correlation score dropped from 0.51 to 0.42 (p < 0.0001 for correlation decrease). By eliminating the SVR step and using the average thickness of each AAL region as an input to PLS-based domain adaptation, the average correlation score decreased to 0.20 (p < 0.0001 for the correlation decrease). Using elastic net directly on regional cortical thickness values yielded the average correlation score of 0.05.
Supporting Image: scatter_diffmethods_OHBM2017.png
   ·Figure 2. Scatter plot between observed and estimated severity scores.


The proposed method for severity score prediction performed better than simpler alternatives and resulted in a higher cross-validated correlation score than has previously been reported (Sato 2013, for raw ADOS scores and standardized acquisition). The utility of the proposed approach for the multi-site, multi-protocol ABIDE dataset indicates the potential of designing machine learning methods to meet the challenges of agglomerative data.

Disorders of the Nervous System:

Autism 2

Imaging Methods:

Anatomical MRI

Modeling and Analysis Methods:

Classification and Predictive Modeling 1

Poster Session:

Poster Session - Wednesday


Data analysis
Machine Learning

1|2Indicates the priority used for review

Would you accept an oral presentation if your abstract is selected for an oral session?


I would be willing to discuss my abstract with members of the press should my abstract be marked newsworthy:


Please indicate below if your study was a "resting state" or "task-activation” study.


By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute the presentation in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels or other electronic media and on the OHBM website.

I accept

Healthy subjects only or patients (note that patient studies may also involve healthy subjects):


Internal Review Board (IRB) or Animal Use and Care Committee (AUCC) Approval. Please indicate approval below. Please note: Failure to have IRB or AUCC approval, if applicable will lead to automatic rejection of abstract.

Not applicable

Please indicate which methods were used in your research:

Structural MRI

For human MRI, what field strength scanner do you use?


Which processing packages did you use for your study?

Other, Please list  -   CIVET

Provide references in author date format

Di Martino, A., Yan, C.G., Li, Q., Denio, E., Castellanos, F.X., Alaerts, K., Anderson, J.S., Assaf, M., Bookheimer, S.Y., Dapretto, M., et al., 2014. The autism brain imaging data exchange: towards a large-scale evaluation of the intrinsic brain architecture in autism. Molecular psychiatry 19, 659–667.

E. Moradi, B Khundrakpam, J. Lewis, A.C. Evans, J. Tohka (2016). Predicting symptom severity in autism spectrum disorder based on cortical thickness measures in agglomerative data, Neuroimage, in press

Sato, J.R., Hoexter, M.Q., de Magalh˜aes Oliveira, P.P., Brammer, M.J., Murphy, D., Ecker, C., et al., 2013. Inter-regional cortical thickness correlations are associated with autistic symptoms: a machine-learning approach. Journal of psychiatric research 47, 453–459.