Development of a PDA-based wearable digital phonocardiograph

In this thesis the applicability of advanced digital signal processing techniques to the analysis of heart sound is first demonstrated. The development of a PDA-based phonocardiograph with heart sounds signal acquisition, processing and analysis is then described. Fourier transform-based spectral analysis of heart sounds was carried out to first show the differences in the frequency contents of normal and abnormal heart sounds. However, the time-varying nature of heart sounds called for better techniques capable of analyzing such signals. Therefore, two more techniques namely: the Short-time Fourier Transform (STFT) or spectrogram and the Wavelet Transform (WT) analysis were deployed. These methods performed remarkably well in displaying frequency, magnitude, and time information of the heart sounds, providing robust parameters to make accurate computer-aided classification. These feature vectors were fed into two Artificial Neural Networks (ANNs). The Multilayer Perceptron (MLP) architecture with the backpropagation learning algorithm was selected for these ANNs. The first network served as a general classifier and the second as a classifier for particular (specific) diseases. Then a number of normal and abnormal heart sounds were classified to validate the effectiveness of the statistical segmentation and the feature extraction methods. The final results demonstrated that the general classification yielded 84% accuracy for the PC version and a accuracy 70% accuracy for PDA version, respectively. The specific classification yielded an accuracy of 68% for the PC version and 50% for the PDA version, respectively. Future research will improve these performance levels to better than 95% accuracy for the PDA version. (Abstract shortened by UMI.