Psychiatric disorders from EEG signals through deep learning models.

Area of Science:

• Neuroscience
• Psychiatry
• Computer Science

Background:

• Psychiatric disorder diagnosis is challenging due to emotional concealment and limitations of traditional neurophysiological methods.
• Electroencephalography (EEG) offers a potential avenue for objective diagnostic measures.

Purpose of the Study:

• To develop and evaluate an improved EEG-based diagnostic model using Deep Learning (DL) techniques for psychiatric disorders.
• To enhance the accuracy and reliability of psychiatric disorder diagnosis.

Main Methods:

• Utilized a dataset of 945 individuals (850 patients, 95 healthy controls) focusing on six main and nine specific psychiatric disorders.
• Analyzed quantitative EEG data (resting state) including power spectral density (PSD) and functional connectivity (FC) across frequency bands.
• Employed and compared various DL models: Artificial Neural Networks (ANN), K-Nearest Neighbors (KNN), Long Short-Term Memory (LSTM), Bidirectional LSTM (Bi-LSTM), and CNN-LSTM for binary classification.

Main Results:

• All proposed DL models demonstrated superior performance compared to previous methods.
• ANN achieved 96.83% accuracy for obsessive-compulsive disorder (OCD); CNN-LSTM achieved 96.83% for adjustment disorder.
• KNN and LSTM reached 98.94% accuracy for acute stress disorder; KNN and Bi-LSTM achieved 97.88% for OCD prediction.

Conclusions:

• EEG, enhanced by DL, shows significant potential as a cost-effective and accessible diagnostic tool for psychiatric disorders, complementing methods like MRI.
• Advanced DL models applied to EEG data can improve the detection, monitoring, and clinical application of psychiatric disorder diagnosis.
• This approach holds promise for improved patient care and outcomes in psychiatry.