Classification of Doppler Blood Flow Sound During Hemorrhoidal Artery Ligation Using Mel Frequency Cepstrum Coefficient and Support Vector Machine

In 1995, Morinaga et al. adopted the Doppler blood flow measurement technique to develop a novel surgical treatment for haemorrhoids, later known as the Doppler guided hemorrhoidal artery ligation (DG-HAL). During an artery search process, the surgeon is required to focus on the ultrasound produced by the Doppler instrument to perceive subtle changes indicating the presence of a target artery. However, the correct identification of haermorrhoidal arteries using ultrasound may sometimes be difficult because of several factors. The use of automated methods as a decision support system can help the DG-HAL practitioners in identifying hemorrhoidal arteries in better ways. However, the Doppler audio signal is not frequently recorded and investigated in the literature despite advancements in biological sound classification in recent years. Therefore, this paper proposes a Doppler blood flow sound classification method based on Mel frequency cepstrum coefficient (MFCC) features and support vector machine (SVM) classifier. Several MFCC input combinations to SVM, kernel functions, and SVM parameters are investigated to determine the best accuracy. The experiment results indicate that for cases of binary classification (arterial sound and non-arterial sound), a classification accuracy of 96.66% has been reached on a private database consisting of 100 sound samples.