Causal Inference of ASD diagnosis from rest-fMRI Connectivity and Confounders using Machine Learning
Poster No:
363
Submission Type:
Late-Breaking Abstract Submission
Authors:
Melanie Garcia1, Benjamin Wade2, Joan Camprodon2
Institutions:
1Mass. General Hospital, Harvard Medical School, Charlestown, MA, 2Massachusetts General Hospital, Charlestown, MA
First Author:
Co-Author(s):
Introduction:
This study proposes a modelling approach that accounts for confounding factors to build functional connectivity subtypes of Autism using machine learning.
Autism Spectrum Disorder is heterogeneous. Many studies have shown the importance of studying functional connectivity (Horien et al., 2022; Traut et al., 2021), but connectivity alone does not fully explain the heterogeneity of ASD (Horien et al. 2022). Confounding factors can also influence connectivity properties and impact ASD diagnosis.
Causal modelling (Imbens, G. W. and Rubin, D. B., 2015) provides a rigorous framework to incorporate confounding factors such as age, gender, and data acquisition protocols, which affect functional connectivity. This approach enables a more refined analysis of heterogeneous subgroups and offers direct interpretability at both the connectivity level and the diagnostic level for the various subgroups.
Previous studies using resting-state fMRI data have been descriptive or have applied machine learning to predict ASD directly from functional connectivity, without integrating a causal modelling framework to disentangle heterogeneous subgroups (Horien et al., 2022).
Autism Spectrum Disorder is heterogeneous. Many studies have shown the importance of studying functional connectivity (Horien et al., 2022; Traut et al., 2021), but connectivity alone does not fully explain the heterogeneity of ASD (Horien et al. 2022). Confounding factors can also influence connectivity properties and impact ASD diagnosis.
Causal modelling (Imbens, G. W. and Rubin, D. B., 2015) provides a rigorous framework to incorporate confounding factors such as age, gender, and data acquisition protocols, which affect functional connectivity. This approach enables a more refined analysis of heterogeneous subgroups and offers direct interpretability at both the connectivity level and the diagnostic level for the various subgroups.
Previous studies using resting-state fMRI data have been descriptive or have applied machine learning to predict ASD directly from functional connectivity, without integrating a causal modelling framework to disentangle heterogeneous subgroups (Horien et al., 2022).
Methods:
Data from ABIDE 1 (Di Martino et al., 2014) was used, excluding scans with poor quality, comorbidities, or medication use, resulting in 773 observations.
fMRI data was preprocessed with C-PAC pipeline, including band-pass filtering (0.01-0.1 Hz), GSR, and registration to MNI152. Mean time series were extracted using the CC200 atlas, and functional connectivity was computed within and between Yeo 7 networks using Pearson correlations.
ASD diagnosis was assumed here to be influenced by extreme connectivity. Connectivity values were binarized into low and high (25% and 75% quantiles), and separate causal models were built for each network or network pair, excluding observations with average connectivity.
A DR-learner approach with 3-fold cross-validation was applied, using elastic net logistic regression for propensity scores and ridge regression for outcome estimation.
For each network or pair, Conditional Average Treatment Effects (CATE) and confidence intervals were estimated, indicating whether high connectivity increased or decreased the likelihood of ASD, or if the effect was uncertain. Permutation tests assessed the role of confounders, and CATE values (positive, negative, or close to zero) were used to identify heterogeneous ASD subgroups.
fMRI data was preprocessed with C-PAC pipeline, including band-pass filtering (0.01-0.1 Hz), GSR, and registration to MNI152. Mean time series were extracted using the CC200 atlas, and functional connectivity was computed within and between Yeo 7 networks using Pearson correlations.
ASD diagnosis was assumed here to be influenced by extreme connectivity. Connectivity values were binarized into low and high (25% and 75% quantiles), and separate causal models were built for each network or network pair, excluding observations with average connectivity.
A DR-learner approach with 3-fold cross-validation was applied, using elastic net logistic regression for propensity scores and ridge regression for outcome estimation.
For each network or pair, Conditional Average Treatment Effects (CATE) and confidence intervals were estimated, indicating whether high connectivity increased or decreased the likelihood of ASD, or if the effect was uncertain. Permutation tests assessed the role of confounders, and CATE values (positive, negative, or close to zero) were used to identify heterogeneous ASD subgroups.
Results:
High connectivity in certain networks or network pairs was more often associated with ASD, while low connectivity was linked to no ASD in other cases.
Overall:
• High connectivity within limbic networks and between default-salience, default-control, default-dorsal attention, and salience-limbic networks was associated with ASD.
• High connectivity within the default and somatomotor networks, and between visual-control and salience-control networks, was linked to no ASD.
• Certain patterns, such as control-visual-default connectivity, highlighted distinct subgroups.
No single set of confounding variables explained the effect of high connectivity across all networks. Each network or pair had its own key variables. For example, age was important for control network connectivity but added noise for dorsal attention.
By examining the types of estimated CATE values (positive, negative, or uncertain), some confounding factors were linked to specific outcomes. For example, the subgroup where high limbic connectivity was associated with ASD tended to be younger and included more girls than the subgroup where high limbic connectivity was linked to no ASD.
Figure 1 shows results for the default network connectivity model.
Overall:
• High connectivity within limbic networks and between default-salience, default-control, default-dorsal attention, and salience-limbic networks was associated with ASD.
• High connectivity within the default and somatomotor networks, and between visual-control and salience-control networks, was linked to no ASD.
• Certain patterns, such as control-visual-default connectivity, highlighted distinct subgroups.
No single set of confounding variables explained the effect of high connectivity across all networks. Each network or pair had its own key variables. For example, age was important for control network connectivity but added noise for dorsal attention.
By examining the types of estimated CATE values (positive, negative, or uncertain), some confounding factors were linked to specific outcomes. For example, the subgroup where high limbic connectivity was associated with ASD tended to be younger and included more girls than the subgroup where high limbic connectivity was linked to no ASD.
Figure 1 shows results for the default network connectivity model.
·Figure 1
Conclusions:
This study used causal modelling to better capture ASD heterogeneity by accounting for confounders. Connectivity was simplified into binary categories, limiting detail, and interactions between networks were not examined. Future work will model continuous connectivity, and explore measures like mutual information for richer insights.
Disorders of the Nervous System:
Neurodevelopmental/ Early Life (eg. ADHD, autism) 1
Modeling and Analysis Methods:
Classification and Predictive Modeling
Connectivity (eg. functional, effective, structural)
fMRI Connectivity and Network Modeling
Methods Development 2
Multivariate Approaches
Other Methods
Novel Imaging Acquisition Methods:
BOLD fMRI
Keywords:
Autism
Computational Neuroscience
FUNCTIONAL MRI
Machine Learning
MRI
Statistical Methods
1|2Indicates the priority used for review
Abstract Information
By submitting your proposal, you grant permission for the Organization for Human Brain Mapping (OHBM) to distribute your work in any format, including video, audio print and electronic text through OHBM OnDemand, social media channels, the OHBM website, or other electronic publications and media.
The Open Science Special Interest Group (OSSIG) is introducing a reproducibility challenge for OHBM 2025. This new initiative aims to enhance the reproducibility of scientific results and foster collaborations between labs. Teams will consist of a “source” party and a “reproducing” party, and will be evaluated on the success of their replication, the openness of the source work, and additional deliverables. Click here for more information. Propose your OHBM abstract(s) as source work for future OHBM meetings by selecting one of the following options:
Please indicate below if your study was a "resting state" or "task-activation” study.
Healthy subjects only or patients (note that patient studies may also involve healthy subjects):
Was this research conducted in the United States?
Are you Internal Review Board (IRB) certified? Please note: Failure to have IRB, if applicable will lead to automatic rejection of abstract.
Were any human subjects research approved by the relevant Institutional Review Board or ethics panel? NOTE: Any human subjects studies without IRB approval will be automatically rejected.
Were any animal research approved by the relevant IACUC or other animal research panel? NOTE: Any animal studies without IACUC approval will be automatically rejected.
Please indicate which methods were used in your research:
For human MRI, what field strength scanner do you use?
Which processing packages did you use for your study?
Provide references using APA citation style.
Horien, C., Floris, D. L., Greene, A. S., Noble, S., Rolison, M., Tejavibulya, L., O’Connor, D., McPartland, J. C., Scheinost, D., Chawarska, K., Lake, E. M. R., & Constable, R. T. (2022). Functional Connectome–Based Predictive Modeling in Autism. Biological Psychiatry. https://doi.org/10.1016/j.biopsych.2022.04.008
G. W. Imbens and D. B. Rubin. (2015). Causal inference in statistics, social, and biomedical sciences. Cambridge university press.
Traut, N., Heuer, K., Lemaître, G., Beggiato, A., Germanaud, D., Elmaleh, M., Bethegnies, A., Bonnasse-Gahot, L., Cai, W., Chambon, S., Cliquet, F., Ghriss, A., Guigui, N., de Pierrefeu, A., Wang, M., Zantedeschi, V., Boucaud, A., van den Bossche, J., Kegl, B., … Varoquaux, G. (2021). Insights from an autism imaging biomarker challenge: Promises and threats to biomarker discovery [Preprint]. Radiology and Imaging. https://doi.org/10.1101/2021.11.24.21266768