Connectome-based schizophrenia prediction using structural connectivity - Deep Graph Neural Network(sc-DGNN).

Area of Science:

• Neuroscience
• Brain Connectivity
• Machine Learning in Medicine

Background:

• Understanding human brain connectome organization is crucial for insights into cognitive processes and disorders.
• Brain structural and functional connectivity analysis aids in understanding neurological conditions.

Purpose of the Study:

• To enhance prediction accuracy for brain disorder issues, specifically schizophrenia.
• Investigate dysconnected subnetworks and graph structures associated with schizophrenia.

Main Methods:

• Utilized diffusion magnetic resonance imaging (dMRI) data from eighty-eight subjects.
• Developed and applied a structural connectivity-deep graph neural network (sc-DGNN) model.
• Compared sc-DGNN performance against three classical machine learning (ML) and five deep learning (DL) models.

Main Results:

• The sc-DGNN model demonstrated superior predictive performance for schizophrenia-associated dysconnectedness.
• Achieved higher accuracy, sensitivity, specificity, precision, F1-score, and Area Under the ROC Curve (AUC) compared to ML and DL methods.
• sc-DGNN achieved a 93% accuracy rate, significantly outperforming linear discriminant analysis (LDA) at 72% accuracy.

Conclusions:

• The proposed sc-DGNN model effectively distinguishes between schizophrenia patients and healthy individuals.
• Deep graph neural networks offer a promising avenue for improving diagnostic accuracy in psychiatric disorders.
• Structural connectivity analysis combined with advanced DL models holds significant potential for brain disorder research.