End-to-End ECG Signal Compression Based on Temporal Information and Residual Compensation

Multi-lead wearable devices face significant challenges related to data volume, making effective Electrocardiogram (ECG) signal compression essential for efficient transmission and storage. Current methods frequently overlook the inherent temporal correlations of ECG signals, reducing compression effectiveness. This paper proposes a novel ECG signal compression model that optimally balances the compression ratio and signal quality, demonstrating superior performance by leveraging temporal correlations and residual compensation techniques. The proposed model consists of three modules: The inter-frame prediction module reduces data redundancy by utilizing the temporal correlations. The quantization and entropy encoding module adjusts quantization levels based on probability estimates, resulting in diverse compression ratios. The back projection residual compensation module employs back projection to reproject low-dimensional residuals back into the original signal space, thereby mitigating errors during both the compression and reconstruction stages. Extensive validation of the proposed model was conducted using the public datasets CPSC2018, LSEDB-AS and MIT-BIH. Results demonstrate state-of-the-art performance and strong generalization capabilities. The model achieves an impressive average Quality Score score of 25 on the CPSC2018 dataset, offering a Compression Ratio up to 200:1. This research introduces a novel and optimized ECG compression method leveraging temporal correlations and residual compensation. Validated across multiple datasets, it achieves superior compression efficiency while maintaining signal integrity, thereby contributing to the advancement of future ECG compression methodologies. This research is significant for enhancing patient care in telehealth scenarios and addresses storage limitations in healthcare systems, improving overall efficiency and effectiveness.