We present a comparative study for correlation coefficients of three different, but popular transforms of audio signals i.e. Fast Fourier Transform (FFT), Cepstrum and Discrete Cosine Transform (DCT). But this study is done keeping into mind an important application, the heart sound analysis or heart auscultation analysis which manually is done by doctors for disease identification. We present a very simple automated software based approach for first detecting whether the heart is normal or abnormal and then identifying the disease if within the range of diseases for which it has been trained. Here our application has been trained for only three heart diseases, Mitral Regurgitation, Mitral Stenosis, and Splits. Further training might enable our application for identifying other diseases as well. We get better the detection accuracy with the increase of training data. We have taken the help of FBS (Frontiers in Bioscience) online data base for heart sounds for this purpose, and have used their .wav format of heart sound files for our analysis. But along with these we also found out that though Cepstrum is a very important transform for speaker recognition and other audio based application, but here in case of heart sound analysis it is not very user friendly for analysis.
CITATION STYLE
Majumder, S., Pal, S., & Dutta, P. K. (2009). A Comparative Study for Disease Identification from Heart Auscultation using FFT, Cepstrum and DCT Correlation Coefficients. In IFMBE Proceedings (Vol. 23, pp. 754–757). https://doi.org/10.1007/978-3-540-92841-6_185
Mendeley helps you to discover research relevant for your work.