Heart murmurs detection techniques

Human heart is one of the most important organ in the body and one of the most common diagnosis for heart diseases includes heart auscultation. However, the ability to distinguish the presence of abnormal heart sounds, otherwise known as heart murmurs, are heavily reliant on the experience of the physician. As technology advances, it is now plausible to record the Phonocardiogram, in other words, heart sounds as a form of digital signal. As like any digital signal processing, features in the heart sounds must first be extracted in order to best distinguish the heart murmurs from the normal heart sounds. In this project, the features that were explored are namely variance, kurtosis and entropy of the signal, after the signals have passed through Discrete Wavelet Transform of up to 6th decomposition level. Certainly, while 36 features were extracted, not all the extracted features when fed into the classifier will improve the accuracy. As such, feature selection techniques, otherwise known as variable selection techniques, such as Principal Component Analysis and Fisher’s Ratio were applied. These techniques helped to improve the accuracy of the classification by transforming the features obtained to a set of principal components and removing the lesser linearly separable features respectively. Support Vector Machine were then used as the classifier in this project to classify the result, and subsequently used to process the accuracy, specificity and the sensitivity of the algorithm. In the proposed method mentioned in the report, the accuracy achieved was at 81.81%, with 90% and 79.74% in specificity and sensitivity respectively, denoting a relatively promising algorithm with low grade I and grade II error.Bachelor of Engineerin