TanhReLU -based convolutional neural networks for MDD classification.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Machine Learning

Background:

• Major Depression Disorder (MDD) diagnosis is challenging due to its complexity.
• Current data-driven electroencephalogram (EEG) analysis for MDD faces issues like gradient vanishing in classification models.
• Overfitting and gradient vanishing hinder the performance of deep learning models in psychiatric disorder detection.

Purpose of the Study:

• To introduce a TanhReLU-based Convolutional Neural Network (CNN) for improved Major Depression Disorder (MDD) classification using EEG data.
• To mitigate the gradient vanishing problem in deep learning models applied to EEG-based MDD detection.
• To enhance the accuracy and robustness of MDD classification by reducing model overfitting.

Main Methods:

• Development of a novel CNN architecture incorporating the TanhReLU activation function.
• Integration of TanhReLU, which combines Tanh and ReLU characteristics, to improve gradient flow.
• Training and evaluation of the model on publicly available electroencephalogram (EEG) datasets for Major Depression Disorder (MDD) classification.

Main Results:

• The TanhReLU-based CNN demonstrated promising performance in classifying Major Depression Disorder (MDD) from EEG data.
• The model successfully alleviated the issues of gradient vanishing and overfitting.
• Experimental results indicate a significant improvement in classification accuracy compared to existing methods.

Conclusions:

• The TanhReLU-based CNN is an effective approach for EEG-based Major Depression Disorder (MDD) classification.
• This novel activation function offers a viable solution to gradient vanishing and overfitting in deep learning for psychiatric diagnostics.
• The findings suggest potential for clinical application in the objective diagnosis of MDD.