Wheelchair movement signal classification from EEG for motor-impaired individuals using novel deep learning architecture.

Purpose: Traditional wheelchair controls often limit independence and pose safety risks for motor-impaired users. To address these challenges, this study explores the potential of EEG-based control systems that allow users to operate powered wheelchairs through brain signals rather than physical movements. Materials and Methods: We developed a hybrid deep learning model that integrates Long Short-Term Memory (LSTM) and 1D-Convolutional Neural Networks (1D-CNN) with skip connections to capture both temporal and spatial EEG signal features. The model was trained and evaluated on a public EEG dataset to classify intended wheelchair movements. Performance metrics, including accuracy, precision, recall, and F1 score, were computed. Confidence interval tests and ablation studies were conducted to assess statistical reliability and component contribution. Results: The proposed model achieved an accuracy of 98.08% with 0.98 precision, recall, and F1 score, outperforming ten state-of-the-art methods. Confidence interval analysis confirmed the model's statistical superiority, while ablation results demonstrated the importance of the LSTM-CNN fusion and skip connections in enhancing prediction performance. Conclusions: The LSTM-CNN architecture with skip connections offers a reliable and accurate EEG-based control approach for powered wheelchairs, improving safety and independence for users with severe motor impairments. In future, proposed model may lead to EEG-responsive wheelchair systems to aid mobility and self-directed activity, contributing to improved quality of life and rehabilitation outcomes.