Deriving reproducible biomarkers from multi-site resting-state data: An Autism-based example.
Alexandre Abraham1, Michael P Milham2, Adriana Di Martino3
1Parietal Team, Saclay-INRIA le-de-France,Saclay,France; CEA, Neurospin bât 145, 91191 Gif-Sur-Yvette, France.
Neuroimage
|November 21, 2016
Summary
This study shows resting-state functional MRI can identify autism biomarkers across different data sites. Optimized analysis pipelines achieved 67% accuracy, improving autism diagnosis prediction.
Area of Science:
- Neuroscience
- Neuroimaging
- Psychiatry
Background:
- Resting-state functional Magnetic Resonance Imaging (R-fMRI) shows potential for identifying neuropsychiatric biomarkers.
- Autism spectrum disorders present complex challenges for biomarker extraction due to multi-faceted neuropathologies.
- Large, multi-site datasets like the Autism Brain Imaging Data Exchange (ABIDE) offer increased sample sizes but introduce heterogeneity.
Purpose of the Study:
- To demonstrate the feasibility of inter-site classification of neuropsychiatric status using R-fMRI data.
- To investigate pipelines for extracting predictive biomarkers from R-fMRI data for autism.
- To improve prediction accuracy for differentiating individuals with autism from typical controls.
Main Methods:
- Utilized the large, multi-site ABIDE database (N=871) for autism research.
- Developed and investigated R-fMRI pipelines to build participant-specific brain connectomes from functionally-defined areas.
- Compared connectome patterns to learn differentiating connectivity features between controls and individuals with autism.
- Performed inter-site classification, predicting status for participants from both familiar and unseen acquisition sites.
Main Results:
- Achieved 67% prediction accuracy on the full ABIDE dataset, surpassing previous results.
- Demonstrated that prediction accuracy increases with larger sample sizes.
- Found that the definition of functional brain areas significantly impacts biomarker discovery, with data-driven areas outperforming reference atlases.
- Validated findings across multiple data subsets defined by varying inclusion criteria.
Conclusions:
- Inter-site classification of neuropsychiatric status using R-fMRI is feasible, even with data heterogeneity.
- Optimized R-fMRI analysis pipelines can yield robust biomarkers for complex disorders like autism.
- The methodology provides a promising approach for developing more accurate and generalizable diagnostic tools for autism spectrum disorder.
