Predicting Cardiovascular Diseases through Integrated Deep Learning Features

Abstract

This research investigates the effectiveness of various machine learning models for classifying multimodal heart signals, specifically using electrocardiogram (ECG) and phonocardiogram (PCG) data from the EPHNOGRAM and PhysioNet Cardiology 2016 datasets. The proposed architecture, particularly the hybrid (LSTM, BiLSTM) model, achieved an impressive accuracy of 92.3% on the PhysioNet dataset, surpassing existing techniques and demonstrating significant potential for clinical applications. This study evaluates several architectures, including CNN, RNN, CNN and hybrid(RNN, LSTM), BiLSTM, and hybrid(LSTM, BiLSTM). By leveraging the EPHNOGRAM dataset, which contains synchronized ECG and PCG signals, the model aims to enhance classification accuracy for various heart conditions, including normal heartbeats, arrhythmias, murmurs, and other cardiovascular diseases (CVDs). Experimental results indicate that the proposed model outperforms traditional deep learning methods, achieving superior classification accuracy. These findings underscore the promise of multimodal deep learning models in healthcare diagnostics, particularly in detecting cardiac diseases. Future research will focus on optimizing hyperparameters, utilizing larger datasets, and integrating additional modalities, such as cardiac imaging and demographic data, to improve cardiovascular disease classification further.