AN EFFECTIVE DEEP LEARNING BASED APPROACH TO MEASURE INTERDEPENDENCE OF HEART RATE VARIABILITY ANALYSIS AND R-PEAKS

Abstract

Heart rate variability (HRV) is the physiological phenomenon to measure variations between consecutive heartbeats. HRV analysis has been used to analyse and detect different cardiac diseases that rely on R-peaks' reliability and accurate detection. Myocardial infarction (MI) is a main cardiac disease that causes irregularity in heartbeats and some non-specific abnormalities occur in the recorded ECG, including ST changes and T-wave alternans. These abnormalities may be ignored by considering minor changes that can lead to delayed diagnosis. Therefore, the analysis of these abnormalities is a challenging task. This study investigated these non-specific abnormalities to recognize MI patterns based on HRV analysis highlighting the importance of R-peaks detection accuracy. Short-term HRV analysis of two publicly available datasets, MIT-BIH ST Changes and European ST-T changes was performed based on R-peaks detected from ECG signals. R-peaks detection achieved an averaged 99.77% sensitivity, 99.49% positive predictability and 99.32% accuracy for MIT-BIH ST changes dataset whereas for European ST-T changes dataset, 99.85% sensitivity, 99.41% positive predictability, and 99.28% accuracy was achieved. Subsequently, HRV parameters were computed from both datasets and data fusion was performed followed by a deep learning model to find out the most accurate pattern recognition results. For pattern recognition, a finely tuned artificial neural network was applied to HRV parameters, in two scenarios. For both scenarios, an accuracy greater than 99% was achieved. Furthermore, linear regression was also performed on computed HRV measures in both scenarios that observed a similarity of 97% between both datasets. The importance of reliable and accurate R-peaks detection for HRV analysis to analyse and detect different cardiovascular diseases was elaborated in the end.