A deep learning framework for automatic diagnosis of unipolar depression.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Medical Diagnostics

Background:

• Machine learning frameworks for depression diagnosis are advancing rapidly, particularly deep learning.
• Further validation is needed for these advanced ML models in clinical settings.

Purpose of the Study:

• To propose and validate an electroencephalographic (EEG)-based deep learning framework for automatic diagnosis of unipolar depression.
• To discriminate between individuals with depression and healthy controls using EEG data.

Main Methods:

• Development of two deep learning architectures: one-dimensional convolutional neural network (1DCNN) and a hybrid 1DCNN with long short-term memory (LSTM).
• Utilizing resting-state EEG data from 33 patients diagnosed with depression and 30 healthy controls for model validation.
• Automatic pattern learning from EEG data for classification of depressed and healthy subjects.

Main Results:

• The 1DCNN model achieved high classification performance: 98.32% accuracy, 99.78% precision, 98.34% recall, and 97.65% f-score.
• The 1DCNN-LSTM model demonstrated strong results with 95.97% accuracy, 99.23% precision, 93.67% recall, and 95.14% f-score.
• Significant differences were identified between the depressed and healthy control groups based on EEG patterns.

Conclusions:

• Deep learning frameworks show potential to significantly advance clinical applications for EEG-based depression diagnosis.
• The proposed deep learning framework can serve as an effective automatic method for diagnosing depression.