Performance Analysis of GMM Classifier for Classification of Normal and Abnormal Segments in PPG Signals

Abstract

Photoplethysmography (PPG) is widely used to estimate the blood flow of skin by utilizing the infrared technique. Parameters such as blood pressure, oxygen saturation levels, blood saturation levels and cardiac output levels can be measured easily. As PPG is non-invasive in nature and it has a low production and maintenance cost, it is widely used in clinical practices. The performance analysis of Gaussian Mixture Model (GMM) as a post classifier is utilized in this paper for the classification of normal and abnormal segments in PPG Signals. The main objective of the paper is to identify normal and abnormal PPG Segments of the PPG waveform observed in the long time monitoring of the Physionet Data Base available online for a particular patient. The PPG Signals are sampled at 100 Hz. The PPG data sample length obtained is 1, 44, 000 and it is segmented into equal intervals comprising of 200 samples totally. Therefore the entire data consists of 720 segments. Totally ten different features such as mean, variance, standard deviation, skewness, kurtosis, energy, approximate entropy, peak maximum, maximum slope, and Singular Value Decomposition (SVD) are extracted and normalized. Based on the SVD values, each segment is labeled as normal or abnormal segment. The normalized features are given as inputs to the GMM classifier to classify the normal and abnormal segments in the PPG Signals. The performance metrics analyzed in this work are specificity, sensitivity, accuracy, precision and False Discovery Rate (FDR). Results show that an accuracy of 98.97% is obtained, precision of 100%, nil FDR, specificity of 100% and sensitivity of 97.95% is obtained.