Cardiovascular Condition Monitoring based on Heart Murmur Analysis

Main Article Content

Ravindra Manohar Potdar, Mekh Ram Meshram, Ramesh Kumar

Abstract

Usage of classical auscultation techniques is qualitative and subjective. The decision making process entirely depends upon the expertise and experience of the physicians conducting the observations. In order to make the outcome of the observations quantitative and independent of expertise and experience of the observer, Automated Computerized Analysis (ACA) of heart murmurs is best suited. Efficacy of classification of heart murmurs largely depends upon the feature extraction process. Murmurs, being a non-stationery signal, features in the time domain, frequency domain, spatial domain and phase domain are necessarily to be extracted. Such features will act as input to the classifier based on machine learning. In the present work, a new technique for extraction of features of heart murmurs is proposed to enhance its suitability for classification. Features were extracted in the time domain, Frequency domain, Time-Frequency domain and Phase space domain. As the size of the feature vectors were found to be very large, feature dimension reduction method was applied to reduce the number of feature vectors by eliminating redundant or dependent features. These features were then applied as input to the classifier for murmur screening. Four types of murmurs namely musical quality, coarse quality, soft quality and blowing like quality were considered for feature extraction. The heart signal data were obtained from open sources. PASCAL Heart Sound challenge Data set 2011 [1].


Various techniques were applied at different stages. DWT was employed for denoising, feature extraction and dimensionality reduction purposes after segmentation using energy envelogram technique in combination with Gaussian Smoothing Filter (GSF). Support Vector Machine (SVM) with linear kernel function, Linear Discriminant Analysis (LDA), K-Nearest Neighbors (kNN) Algorithm, Probabilistic Neural Network (PNN) techniques were applied for classification of the feature vectors to draw the conclusion regarding the screening of the murmurs. The present work confirmed that it is feasible to screen heart murmurs as normal or abnormal effectively using the techniques stated. There are scopes for further research work in the same direction to discriminate the causes of abnormalities present in the heart.

Article Details

Section
Articles