Federated learning-driven quantum models for EEG-based neurological disorder classifications

Electroencephalography (EEG) enables non-invasive detection of neurological disorders with Deep Learning (DL). However, traditional methodologies for the training of DL models raises privacy concerns due to its centralized training. Federated learning (FL) mitigates these issues by training models in decentralized manner while keeping data on-device. Most peer studies rely on centralized data, and it remains unclear whether emerging quantum classifiers and circuits can match or surpass DL models under FL settings. This study, therefore, presents FL framework for EEG classification using DL models and quantum models while preserving the data locality. Five clients trained locally on 80/20 splits and shared only local model weights for aggregation of the global model. DL models included EEGNetv1, EEGNetv4, EEGITNet, and EEGInception. Quantum models included Quantum Support Vector Machine (QSVM), Variational Quantum Classifier (VQC), and Quantum Convolutional Neural Network (QCNN). Two public datasets e.g., OpenNeuro and TUH EEG Corpus have been utilized. Results show that on the OpenNeuro dataset QSVM global model outperformed with a classification accuracy of 93.57%, and among DL models EEGNetv1 outperformed with an accuracy of 96.92%. Whereas, for the TUH dataset the QCNN model outperformed with an accuracy of 94.52% and EEGInception outperformed among DL models with an accuracy of 92.38%, respectively. This framework demonstrates decentralized training under FL settings and shows that quantum models are comparable to tailored EEG DL models for multiclass classification problem. The results demonstrate the feasibility of integrating FL with quantum models for decentralized EEG based diagnosis, highlighting the practical approach for healthcare industry.