Phonocardiogram classification using time-frequency representation

Introduces a method for classifying systolic murmurs using the time-space representation of a phonocardiogram (PCG). The method detects the presence of systolic murmurs by a neural network utilizing variance sequences. Then, a wavelet transform is applied to that data classified as being related to a systolic murmur. A second neural network classifies the time-frequency distribution thus obtained into abnormal and benign murmurs. The proposed method was applied to data obtained from a nationwide phonocardiogram screening of elementary school children conducted in Japan. A correct decision rate of 98% was achieved for the first stage (detecting the presence of a systolic murmur). Correct classification rates for abnormal and benign murmurs were 78.6% and 84.5% respectively. The method could be useful in assisting medical doctors to make a final decision