EEG-Based Surface EMG Reconstruction Using Deep Sequence Learning for Upper Limb Motor Activity

Purva Jain, Varad Bharadiya, Unnath Chittimalla, Yash Anand, Madhav Rao

Confirmation: I have read and agree with the IEEE BHI 2025 conference submission's policy on behalf of myself and my co-authors.

Keywords: EEG, EMG, deep learning, CNN-LSTM, sequence-to-sequence, SoftDTW, cortico-muscular coupling, neuro-technology

Abstract: We propose a novel EEG-based approach for reconstructing surface EMG signals using deep sequence learning, enabling accurate decoding of muscle activity without requiring direct muscular sensing. This paper presents a foundational proof-of-concept, establishing the feasibility of using a deep learning framework to learn direct cortico-muscular mappings from non-invasive EEG. The framework integrates advanced signal preprocessing with spatio-temporal modeling to achieve this goal. By replacing traditional dual-modality EEG-EMG systems with a single EEG modality, this method significantly reduces hardware complexity while maintaining high fidelity in neuromuscular decoding. The synthesized surface EMG signal from the trained model closely matches the true EMG signal. The proposed model holds promise for streamlined wearable neurotechnology in assistive control, rehabilitation feedback, and motor intent interpretation.

Track: 1. Biomedical Sensor Informatics

Registration Id: ZZN98QSR73L

Submission Number: 277