NAS-optimized topology-preserving transfer learning for differentiating cortical folding patterns.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Imaging

Background:

• Cortical folding patterns exhibit structural and functional differences.
• Advanced techniques like deep learning are underutilized for analyzing these differences.
• Exploring cortical fold variations can reveal insights into neurological conditions.

Purpose of the Study:

• To develop a topology-preserving transfer learning framework for differentiating functional magnetic resonance imaging (fMRI) time series from cortical folds.
• To investigate structural and functional differences in cortical folding patterns between individuals with autism spectrum disorder (ASD) and healthy controls (HC).
• To assess the potential of these differences as biomarkers for ASD diagnosis.

Main Methods:

• Proposed a framework integrating Neural Architecture Search (NAS) for network design.
• Developed a topology-preserving transfer method (TPNAS-Net) converting 2D operations to 1D for analyzing fMRI time series.
• Applied TPNAS-Net for classification of fMRI data and correlation analysis with age in ASD and HC groups.

Main Results:

• TPNAS-Net achieved high classification accuracy in discriminating cortical folding patterns.
• The framework identified subtle functional differences between ASD and HC groups (p=0.042).
• A positive correlation was found between classification accuracy and age in individuals with ASD (r=0.39, p=0.04).

Conclusions:

• The TPNAS-Net framework effectively differentiates cortical folding patterns, offering a novel approach for analyzing brain function.
• Identified significant differences in cortical folding patterns between ASD and HC individuals.
• Cortical folding pattern differences show potential as a biomarker for ASD diagnosis and may correlate with age.