Multimodal cross-scale context clusters for classification of mental disorders using functional and structural MRI.

Area of Science:

• Neuroscience
• Computational Psychiatry
• Medical Imaging Analysis

Background:

• Mental disorders present diagnostic challenges due to the brain's complex, multi-scale nature.
• Understanding dynamic brain activity across different scales is crucial for identifying neural correlates of psychiatric conditions.

Purpose of the Study:

• To develop a novel computational model for analyzing brain activity across multiple scales.
• To uncover hidden biomarkers associated with mental disorders by modeling dynamic segregation and integration of brain regions.

Main Methods:

• Proposed the multimodal cross-scale context clusters (MCCocs) model, integrating multimodal brain image data.
• Utilized a Voxel Reducer for local feature extraction and dimensionality reduction within regions of interest (ROIs).
• Employed an ROI Context Cluster Block for unsupervised clustering and modeling inter-ROI interactions, simulating dynamic scale switching.

Main Results:

• The MCCocs model successfully integrated multimodal voxel information into a novel brain representation.
• Demonstrated the model's ability to capture local associations and global integration patterns across brain regions.
• Achieved state-of-the-art performance in classifying multiple mental disorders, identifying potential discriminative biomarkers.

Conclusions:

• The MCCocs model offers a powerful approach for analyzing complex brain dynamics in mental disorders.
• Simulating scale switching provides new insights into brain structure-function relationships relevant to psychiatric conditions.
• The model's performance suggests its potential for clinical application in diagnosing mental disorders.