ECG Abnormality Detection from PPG Signal

Abstract

Cardiac Arrhythmia is a serious and common cardiovascular disease. Electrocardiographic signal (ECG) is usually used for diagnosis. In addition, multi model signal is also used to reduce false alarms in detecting abnormalities. New directions were turned around using Photoplethysmographic signal (PPG) because of its tremendous advantages. This paper presents an algorithm to detect heart beat abnormalities from PPG signal only (mono type). This method relies on 11 features extracted from 3 subgroups: 4 features from hemodynamic system frequency response where each PPG beat was assumed to be the system output, 5 features were extracted from fitting technique. In which each PPG pulse was approximated as a mixture of Poisson function. Moreover, 2 features were related to time domain-properties. This algorithm was tested and evaluated using MIMIC and MIMIC III databases. Features were then classified using Random forest tree and k-nearest neighbors (KNN) classifiers, promising results were obtained from both with values of 99%, 95% for accuracy, precision respectively