[Paper Review] Depression Diagnosis and Drug Response Prediction via Recurrent Neural Networks and Transformers Utilizing EEG Signals

Abdolkarim Saeedi, Arash Maghsoudi+1 more|arXiv (Cornell University)|Mar 9, 2023

EEG and Brain-Computer InterfacesNeuroscience3 citations

TL;DR

This study proposes a transformer-based deep learning model that leverages EEG signals to diagnose major depressive disorder (MDD) and predict individual patient response to pharmacological treatment. The method achieves 97.14% accuracy and 99.41% recall in MDD classification, and 97.01% accuracy and 97.76% recall in distinguishing drug responders from non-responders, outperforming CNN, LSTM, and CNN-LSTM baselines with fewer parameters.

ABSTRACT

The Early diagnosis and treatment of depression is essential for effective treatment. Depression, while being one of the most common mental illnesses, is still poorly understood in both research and clinical practice. Among different treatments, drug prescription is widely used, however the drug treatment is not effective for many patients. In this work, we propose a method for major depressive disorder (MDD) diagnosis as well as a method for predicting the drug response in patient with MDD using EEG signals. Method: We employ transformers, which are modified recursive neural networks with novel architecture to evaluate the time dependency of time series effectively. We also compare the model to the well-known deep learning schemes such as CNN, LSTM and CNN-LSTM. Results: The transformer achieves an average recall of 99.41% and accuracy of 97.14% for classifying normal and MDD subjects. Furthermore, the transformer also performed well in classifying responders and non-responders to the drug, resulting in 97.01% accuracy and 97.76% Recall. Conclusion: Outperforming other methods on a similar number of parameters, the suggested technique, as a screening tool, seems to have the potential to assist health care professionals in assessing MDD patients for early diagnosis and treatment. Significance: Analyzing EEG signal analysis using transformers, which have replaced the recursive models as a new structure to examine the time dependence of time series, is the main novelty of this research.

Motivation & Objective

• To develop a reliable, non-invasive method for early diagnosis of major depressive disorder (MDD) using EEG signals.
• To predict individual patient response to antidepressant medication using EEG-based machine learning.
• To evaluate the performance of transformer architectures against traditional deep learning models (CNN, LSTM, CNN-LSTM) in EEG-based MDD and drug response classification.
• To establish a screening tool that supports clinical decision-making in mental health by improving diagnostic accuracy and treatment personalization.

Proposed method

• The study employs a modified transformer architecture designed to effectively model temporal dependencies in EEG time series data.
• The model processes raw or preprocessed EEG signals through multi-head self-attention mechanisms to capture long-range and complex temporal patterns.
• The architecture is trained end-to-end on EEG data from healthy controls and MDD patients, with separate heads for MDD diagnosis and drug response classification.
• Performance is benchmarked against established models: CNN, LSTM, and CNN-LSTM, using identical data splits and evaluation metrics.
• The method uses attention weights to interpret model decisions, enhancing clinical interpretability.
• The model is optimized using standard deep learning loss functions (e.g., cross-entropy) and training protocols.

Experimental results

Research questions

• RQ1 Can a transformer-based model achieve superior performance in classifying MDD patients versus healthy controls using EEG signals compared to traditional RNNs and CNNs?
• RQ2 Can the same EEG-based model accurately predict whether an MDD patient will respond positively to pharmacological treatment?
• RQ3 How does the attention mechanism in transformers contribute to identifying relevant EEG patterns for MDD and drug response prediction?
• RQ4 Does the transformer model maintain high performance with fewer parameters than conventional deep learning architectures?

Key findings

• The transformer model achieved 97.14% accuracy and 99.41% recall in distinguishing MDD patients from healthy controls using EEG signals.
• For drug response prediction, the model reached 97.01% accuracy and 97.76% recall in classifying responders versus non-responders to medication.
• The transformer outperformed CNN, LSTM, and CNN-LSTM models in both diagnostic and response prediction tasks despite using a similar number of parameters.
• The attention mechanism enabled the model to focus on clinically relevant EEG patterns, enhancing interpretability and performance.
• The results demonstrate the potential of transformers as a screening tool for early MDD diagnosis and personalized treatment planning.