A novel dimensionality reduction approach for ECG signal via convolutional denoising autoencoder with LSTM

Typical IoT based e-health scenarios use resource constrained wearable device to facilitate ubiquitous long-term monitoring for chronic conditions like cardiovascular disease (CVD). Electrocardiogram (ECG) is an efficient indicator to diagnose patients with CVD. In wearable technology, the signal transmission cost is high and the observed ECG signal is likely to be contaminated with noise. To amend this, an efficient lightweight signal compression scheme is designed to reduce the signal size before transmitting it and thereby reducing the transmission cost and allow long-term monitoring. In this paper, an edge based novel approach is proposed by combining convolutional denoising autoencoder (CDAE) and long short-term memory (LSTM) for ECG signal compression. A single layered LSTM network is added to the end of encoder section of the CDAE, instead of adding several convolutional filters and pooling layers. In which, the number of trainable parameters of the model are reduced and in turn lessen the computation time. Also, the LSTM network learns the order dependencies between the data that helps to reconstruct the data from its compressed form. In the meantime, the proposed algorithm denoises the signal as it employs Denoising Autoencoder architecture. The experiments are conducted on ECG signal taken from MIT-BIH Arrhythmias Database. The experimental result shows that the proposed method is efficient by achieving compression ratio of 64 with better reconstruction quality score of 15.61 which is higher than state-of-the-art methods. As well the proposed method is lightweight when compared with baseline methods CDAE and stacked autoencoder in terms of computation cost.