A Depression Prediction Algorithm Based on Spatiotemporal Feature of EEG Signal.

Area of Science:

• Neuroscience
• Computational Psychiatry
• Biomedical Engineering

Background:

• Depression is a prevalent global mental disorder, posing diagnostic challenges.
• Current diagnostic methods may lack objectivity and effectiveness.
• There is an urgent need for reliable methods to aid depression diagnosis.

Purpose of the Study:

• To propose a novel strategy for predicting depression using electroencephalogram (EEG) spatiotemporal features.
• To develop an auxiliary diagnostic tool for depression.
• To validate the proposed strategy's accuracy and complexity.

Main Methods:

• Denoising EEG signals and obtaining power spectra for Theta, Alpha, and Beta frequency bands.
• Mapping electrode spatial positions to brainpower spectra using orthogonal projection to create brain maps.
• Superimposing brain maps to create a new map integrating frequency and spatial characteristics.
• Utilizing a Convolutional Neural Network (CNN) and Gated Recurrent Unit (GRU) for sequential feature extraction.

Main Results:

• The strategy achieved 89.63% accuracy on a public EEG dataset.
• The strategy achieved 88.56% accuracy on a private EEG dataset.
• The developed network demonstrated low complexity with only six layers.

Conclusions:

• The proposed spatiotemporal feature-based strategy is credible and effective for depression prediction using EEG signals.
• The method offers a less complex and potentially valuable tool for auxiliary depression diagnosis.
• This approach holds promise for improving the objective identification of depression.