CS-EEG: A Deep Compressed Sensing Approach for Single-Channel EEG Reconstruction

Compressed Sensing (CS) provides effective methods for data compression and reconstruction, emerging as a powerful technique in biomedical signal processing. In this paper, we propose an enhanced CS algorithm named CS-EEG, that is, we combine Orthogonal Matching Pursuit (OMP) with a CNN-LSTM network to improve the reconstruction quality of compressed electroencephalogram (EEG) signals. We applies this method to the Sleep-EDF-SC dataset, specifically using single channel EEG recordings to improve efficiency. The reconstruction quality was evaluated based on reconstruction time (RT) and percentage root-mean-square difference (PRD). Experimental results demonstrate that the CNN-LSTM refinement significantly reduces PRD compared to OMP alone, achieving a PRD of 6.99% vs 7.87% with OMP in 460 Hz with 70% compression rate. This shows that the proposed CS scheme can faithfully reconstruct EEG signals using only 30% of the sampling rate, highlighting the potential of CS combined with deep learning in efficiently processing continuous EEG signals with minimal information loss.