Analysis of Classifier Performance on ECG Interpretation for Precision Medicine: Which performance metrics should we use?

Abstract

In Indonesia, the prevalence of cardiac disease diagnosed by doctors and symptoms higher in rural areas and in the lowest ownership index quintiles. Software engineering based on deep learning techniques makes technology-based updates dynamically evolve for health care, especially in interpreting cardiac abnormalities. Electrocardiogram (ECG) is still one of the reliable cardiac examination tools. ECG is one source of medical big data collected, in addition to electronic health records (EHR), biomarker data, medical imaging, biometric data, etc. Medical big data will be processed for decision making to assign a diagnosis among several possible patients diagnose. The quality of a decision support system to interpret actual patient conditions is determined by how accurately the system is able to represent the diagnosis by experts. Some analytical targets of medical big data are prediction (classification). In classification, the performance classifiers can be evaluated by various performance metrics tested in a set of tests or independent validation sets. Performance metrics are used for helping the researchers interested in biomedical engineering to evaluate the performance of classifiers. Multivariate and imbalanced data are major problems in evaluating the performance. Therefore, in this paper discuss the performance metrics that used to minimize the bias of the classifier performance in ECG for enhancing the quality of diagnoses of cardiac abnormality in rural health care. This paper uses Deep Learning technique to show the results of performance metrics for ECG interpretation in classifying cardiac abnormalities.