• Federated learning framework for privacy-preserving EEG-based MDD detection • Gamma-band EEG modeling captures discriminative depression patterns • Signal-aware 1D-CNN enables stable learning under non-IID federated data • Comparative evaluation of FedAvg, FedNova, and ensemble aggregation • Low latency supports real-time clinical EEG inference and deployment The high prevalence of Major Depressive Disorder (MDD) underscores the need for innovative diagnostic methods that are both accurate and respectful of patient privacy. This study addresses the challenge of accurately predicting MDD from distributed patient data while maintaining data confidentiality. We propose a two-step diagnostic framework: first, it analyzes EEG signals to detect unusual patterns, and then it uses a web-based system leveraging federated learning combined with a Convolutional Neural Network (CNN) for the final prediction of MDD. A key contribution of this research is the introduction of a weighted model averaging algorithm that enhances traditional federated learning by incorporating client-specific loss values, thereby improving model performance. The system is developed using the Django framework, ensuring robustness and scalability. Experimental results show an accuracy of 87.48%, a precision of 81.87%, a recall of 97.90%, and an Area Under the Curve (AUC) of 96.86%. These results demonstrate the potential of our federated learning approach for accurate and private prediction of MDD. By leveraging EEG signals, this approach provides a non-invasive, reliable diagnostic tool for early detection of MDD, with significant potential for clinical practice. The findings of this study contribute to neuropsychiatric research and support the development of privacy-preserving machine learning systems in healthcare.
Parvej et al. (Wed,) studied this question.