Multi-scale convolutional recurrent neural network for psychiatric disorder identification in resting-state EEG.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Psychiatry

Background:

• Accurate classification of psychiatric disorders is crucial for developing individualized treatments.
• Objective neuroimaging biomarkers are needed for reliable diagnosis and monitoring.
• Current diagnostic methods can be subjective and lack objective biomarkers.

Purpose of the Study:

• To investigate a deep learning model, MCRNN, for classifying psychiatric disorders using rsEEG.
• To explore novel biomarkers for schizophrenia, bipolar disorder, and major depressive disorder.
• To leverage spatiotemporal information from rsEEG for enhanced diagnostic accuracy.

Main Methods:

• Utilized a multi-scale convolutional recurrent neural network (MCRNN) for classification.
• Analyzed resting-state EEG (rsEEG) data from 327 individuals (schizophrenia, bipolar, MDD, healthy controls).
• Mapped subjects to a low-dimensional subspace for inter-relationship interpretation.

Main Results:

• Achieved high accuracy in two-class classification (78.6-91.3%) and 68.2% in four-class classification.
• Identified discriminative rsEEG biomarkers for psychiatric disorders.
• The model's control-to-schizophrenia trajectory aligned with clinical observations of disease severity.

Conclusions:

• MCRNN effectively extracts discriminative rsEEG biomarkers for psychiatric disorder classification.
• The model shows potential for advancing the understanding of psychiatric disorders.
• This approach may aid in monitoring treatment interventions for psychiatric conditions.