A novel multichannel sparse convolutional autoencoder for electrocardiogram signal compression

Electrocardiogram (ECG) signal compression is paramount in continuously monitoring cardiac patients, as it reduces data storage and transmission costs. Deep neural networks, particularly autoencoders, offer significant potential for compressing ECG signals by mapping them to lower-dimensional spaces. This paper presents a novel multichannel convolutional autoencoder model designed to compress ECG signals efficiently. The proposed approach encodes the ECG signal into a four-channel lower-dimensional space using a convolutional encoder, which is subsequently reconstructed by a deconvolutional decoder. Unlike traditional autoencoder-based methods, the first channel in the model remains unconstrained, while increasing levels of sparsity constraints are imposed on the remaining channels. Different quantization levels are applied to each channel to optimize compression further, reflecting the varying numerical ranges caused by the sparsity constraints. The quantized channels are then encoded using Huffman coding, resulting in a higher compression ratio. The model's effectiveness is evaluated on a popular benchmark dataset, using normalized percent root mean square difference (PRDN) error and compression ratio as performance metrics. The proposed method achieves an average compression ratio of 20.23:1, with an average PRDN error of 9.86%, demonstrating its capability to compress ECG signals efficiently while maintaining reconstruction accuracy.