A Lossless Electrocardiogram Compression System Based on Dual-Mode Prediction and Error Modeling
Long-term electrocardiogram (ECG) monitoring requires high-ratio lossless compression techniques to reduce data transmission energy and data storage capacity. In this paper, we have proposed a high-ratio ECG compression system with low computational complexity. Firstly, as the morphologies of the ECG change over time, we divide the signal of each heartbeat cycle into two regions. To achieve high prediction accuracy, a 1st order linear predictor and a combination of the template predictor and 3rd order linear predictor are applied in the two regions respectively. Secondly, we introduce a context-based error modeling module to the system, which cancels the statistical bias of the prediction algorithm and further improves the prediction accuracy. Thirdly, we modify the Golomb-Rice encoding algorithm to adaptively encode the prediction errors, while preserving a code space for packaging the information that is necessary for prediction. We evaluate the proposed system by using the MIT-BIH Arrhythmia Database (ARRDB). The experimental results show that with memory requirements as low as 444 to 14556 total variables this system achieves a compression ratio (CR) from 2.975 to 3.040, suggesting that it is highly applicable to both the low-power design and the cloud.
Electrocardiogram, lossless compression, error modeling.