Classification of Phonocardiogram Using an adaptive Fuzzy Inference System

This paper proposes a novel approach for the classification of phonocardiograms based on statistical properties of the PCG signal energy envelograms using fuzzy inference system. Fuzzification of features is done to remove absolute boundaries and assign a degree of association to every segment of the signal with the corresponding heart sound. Since heart sound signals are highly nonstationary, characteristic features of the signal segments are usually fuzzified. Developed Mamdani-type fuzzy inference classifier, helps distinguish between different heart sounds and fuzzy features with great accuracy. First of all, sequences of different features of the envelogram are computed which are then statistically manipulated and used as input to the inference system. Rules for the classification are created and output is computed. Crisp results represent degree of association with the correct heart sound. The developed algorithm is tested on standard databases. Results indicate 97% average accuracy to identify different segments of the PCG signal.