Cardiovascular Disease Detection from Multisource, Multilabel ECG Signalswith Wavelet Scattering Transform, Time-Distributed CNNs, and RNNs

Abstract

The classification of cardiac-abnormality patterns plays a crucial role in the diagnosis and treatment of cardiovascular diseases. With the advent of deep learning techniques, particularly convolutional neural networks (CNNs) and recurrent neural networks (RNNs), there has been a significant advancement in the accuracy and efficiency of cardiovascular disease classification with electrocardiogram (ECG) data. However, with the availability of multitudes of freely available multi-source ECG data today, more attempts are required to develop new models that can handle and perform ell on these datasets simultaneously. In this study, an attempt is made to develop a novel deep learningclassification model with multi-source and multi-label ECG dataset for cardiovascular disease classification. Since ECG signals from multiple sources are of different lengths, standardization is done using wavelet scattering transform.  Wavelet scattering transform provides time-frequency features that are independent of length of an ECG signal. A deep learning model is then used to perform cardiovascular disease detection and classification.The model is ahybrid synergistic architecture that uses CNNs in time-distributed fashion and RNNs in bidirectional many-to-many fashionThe CNNs exploit inter-lead correlations and provide temporally compressed features.These temporally compressed featuresare exploited via RNNs to provide a well-generalized model. The multi-source ECG dataset used here is composed of 4 popular 12-lead multi-label ECG datasets available publicly for research purposes.The proposed hybrid model performed satisfactorily overall  on a 27-class classification scenario.