EEG-based Signatures of Schizophrenia, Depression, and Aberrant Aging: A Supervised Machine Learning Investigation.

Area of Science:

• Neuroscience
• Computational Psychiatry
• Biomarker Discovery

Background:

• Electroencephalography (EEG) offers noninvasive, high-temporal-resolution measurement of neuronal activity.
• Machine learning (ML) combined with EEG shows potential for identifying in silico biomarkers for severe mental disorders.

Purpose of the Study:

• To investigate how pathological and physiological aging influence electrophysiological signatures in schizophrenia (SCZ) and major depressive disorder (MDD).
• To develop and validate ML models for classifying SCZ and MDD patients and predicting age using EEG data.

Main Methods:

• Resting-state EEG data were acquired from 735 participants (245 healthy controls, 250 SCZ, 240 MDD).
• Support vector machine models were trained using nested cross-validation to classify diagnostic groups and predict age.
• Electrophysiological Age Gap Estimation (EphysAGE) was calculated as the difference between predicted and chronological age.

Main Results:

• Classification models accurately distinguished SCZ from HC (72.7% BAC), MDD from HC (67.0% BAC), and SCZ from MDD (63.2% BAC).
• Decreased central alpha power was a key predictive feature for both SCZ and MDD.
• Higher EphysAGE correlated with increased likelihood of misclassification as SCZ in healthy controls and MDD patients.

Conclusions:

• ML models can extract distinct electrophysiological signatures for MDD and SCZ, with potential clinical applications.
• The influence of aging on diagnostic discriminability necessitates prompt implementation of these models, particularly for early detection.