Monte Carlo Dropout for Uncertainty Analysis and ECG Trace Image Classification

Md. Farhadul Islam, Sarah Zabeen, Md. Humaion Kabir Mehedi, Shadab Iqbal, Annajiat Alim Rasel

Abstract: Cardiovascular diseases (CVDs), such as arrhythmias (abnormal heartbeats) are the prime cause of mortality across the world. ECG graphs are utilized by cardiologists to indicate any unexpected cardiac activity. Deep Neural Networks (DNN) serve as a highly successful method for classifying ECG images for the purpose of computer-aided diagnosis. However, DNNs can not quantify uncertainty in predictions, as they are incapable of discriminating between anomalous data and training data. Hence, a lack of reliability in automated diagnosis and the potential to cause severe decision-making issues is created, particularly in medical practises. In this paper, we propose an uncertainty-aware ECG classification model where Convolutional Neural Networks (CNN), combined with Monte Carlo Dropout (MCD) is employed to evaluate the uncertainty of the model, providing a more trustworthy process for real-world scenarios. We use ECG images dataset of cardiac and covid-19 patients containing five categories of data, which includes COVID-19 ECG records as well as data from other cardiovascular disorders. Our proposed model achieves 93.90% accuracy using this dataset.