Cardiac Abnormality Detection Model with Inverted Residual Blocks for Large Electrocardiogram Datasets

Abstract

Cardiovascular disease is a critical area of focus in deep learning, as it represents a leading cause of global mortality. Early diagnosis and effective management are crucial in mitigating the impact of heart disease. Deep learning (DL) has demonstrated great promise in revolutionizing various aspects of cardiovascular healthcare. With the ability to process vast amounts of data and recognize intricate patterns, DL models can assist in accurately diagnosing conditions, predicting outcomes, and even suggesting personalized treatment plans. DL models, particularly Convolutional Neural Network (CNN), have shown high accuracy in automatically detecting and classifying cardiac abnormalities. In this work, we propose a new CNN model based on modified inverted residual block (IRB) architecture to perform automatic feature extraction and classification of electrocardiogram (ECG) signals. The research were conducted to identify 2 and 5 distinct classes of heart diseases. The study was performed on data in the PTB-XL dataset. The experimental results demonstrated that our proposed model achieved the highest AUC in each classification task, scoring 94.74% and 93.23% for 2 and 5 classes, respectively. Based on the same public dataset, our results surpass the performance of the state-of the art methods in classifying disease entities into 2 and 5 classes on various evaluation metrics.