Multiple functional networks modeling for autism spectrum disorder diagnosis.

Area of Science:

• Neuroscience
• Medical Imaging
• Machine Learning

Background:

• Autism spectrum disorder (ASD) diagnosis relies on behavioral criteria, lacking definitive neuroimaging biomarkers.
• Resting-state functional magnetic resonance imaging (rsfMRI) shows promise for diagnosing brain disorders like ASD by analyzing functional brain networks.
• Existing rsfMRI methods often depend on predefined models or fail to capture nonlinear, discriminative relationships in functional connectivities (FCs).

Purpose of the Study:

• To propose a novel, data-driven framework for modeling rsfMRI data to improve ASD diagnosis.
• To identify discriminative functional connectivity patterns differentiating individuals with ASD from neurotypical controls.
• To develop a robust classification system for ASD detection using machine learning.

Main Methods:

• Constructing large-scale functional brain networks using hierarchical clustering on rsfMRI data.
• Employing discriminative restricted Boltzmann machines (DRBMs) to learn features and classifiers within identified clusters.
• Utilizing a majority voting strategy based on DRBM outputs for final ASD/NC classification.

Main Results:

• The proposed framework successfully identified discriminative functional connectivity patterns between ASD and normal controls (NC).
• The DRBMs effectively learned features and classifiers, demonstrating strong diagnostic performance on public datasets.
• Comparison with existing methods confirmed the effectiveness and superiority of the novel framework in ASD detection.

Conclusions:

• The developed data-driven framework offers a promising approach for the neuroimaging-based diagnosis of ASD.
• The method accurately distinguishes between ASD and NC by analyzing nonlinear, discriminative functional connectivities.
• Dominant FCs crucial for discriminating ASD were identified, offering insights into the neural underpinnings of the disorder.