Design and characterisation of an automatic stethoscope

Abstract

The work embodied in this dissertation reports the development of an automatic diagnostic system for characterizing phonocardiogram signals obtained using an electronic stethoscope. The use of Hidden Markov Model (HMM) is proposed and implemented for analysis and diagnosis. HMM is a double stochastic process, composed of a stochastic process with an underlying stochastic process which is not observable. Because of HMM's suitability to provide solutions for recognition, segmentation and training problems, it can be used in a predictive statistical heart sound analysis system. There are two core parts to the system: (1) feature extraction, (2) feature recognition. In feature extraction, Mel Frequency Cepstral Coefficients (MFCC) are extracted automatically after filtering and segmentation. Consequently, the feature recognition part builds HMM models according to different heart conditions. With the features extracted in the former part, ten different HMM models can be set up and trained by using left-to-right model, which is a time-based uni-directed model. Each model can denote a particular disease of the heart and a set of models can be determined to represent the conditions of different heart status. The results can then be used for automatic recognition by a probabilistic approach.