Morphometric similarity network-based graph convolutional networks for schizophrenia classification.

Area of Science:

• Neuroimaging
• Computational Psychiatry
• Graph Theory

Background:

• Schizophrenia presents diagnostic challenges due to its heterogeneity.
• Neuroimaging data, particularly brain connectivity, offers potential for classification.
• Existing graph convolutional network (GCN) methods need improvement for subtle schizophrenia-related differences.

Purpose of the Study:

• To introduce a novel GCN framework (MSN-GCN) integrating morphometric similarity networks (MSN) from structural MRI.
• To enhance the classification accuracy of schizophrenia using advanced GCN techniques.
• To identify key brain regions and connectivity patterns associated with schizophrenia.

Main Methods:

• Constructed individual brain graphs using multiple morphometric features (cortical thickness, surface area, etc.).
• Developed a population-level graph incorporating topological and phenotypic information.
• Employed variational edge learning for adaptive optimization of graph edge weights.

Main Results:

• Achieved a superior classification accuracy of 80.85% on a large, multi-site dataset.
• Identified the superior temporal gyrus as a critical region for schizophrenia classification.
• Detected significant differences in clustering coefficients in specific brain regions correlated with negative symptoms.

Conclusions:

• The proposed MSN-GCN framework demonstrates high potential for accurate schizophrenia detection.
• The study provides insights into the neural correlates of schizophrenia through advanced neuroimaging analysis.
• MSN-GCN offers a promising tool for understanding brain structure alterations in neuropsychiatric disorders.