Phonocardiogram classification using self-organizing map with learning vector quantization

This paper proposes a method for classifying the phonocardiogram (PCG) into three target classes (Type I: normal, Type II: innocent murmur, Type III: abnormal murmur). The method detects the presence of systolic murmur by the first self-organizing map with a learning vector quantization algorithm. The second self-organizing map classifies the PCG into innocent and abnormal murmur. The effectiveness of the method was confirmed by applying the method to the data obtained from a nation wide health screening for elementary school children conducted in Japan. The first stage of detecting the presence of systolic murmur achieves a correcting rate of 99.4%, and correcting classification rates for innocent and abnormal murmur were 96.9% and 94.9% respectively. Furthermore, the proposed method not only has good performance for classifying the target classes, but also it could have the ability to classify systolic murmur into sub classes such as ejection murmur or musical murmur, etc. because of the structure of the system