Phonocardiogram Identification Using Mel Frequency and Gammatone Cepstral Coefficients and an Ensemble Learning Classifier

The phonocardiogram, abbreviated as the PCG signal, is one of the signals that has proven to be extremely useful in identifying and diagnosing cardiovascular diseases. Given the ease of acquisition that sets this type of signal apart from others, and knowing that the only tool needed to accomplish the acquisition is a stethoscope, it seems reasonable that identifying clinical signs and symptoms with this signal is extremely valuable. The field of signal processing and artificial intelligence AI has now become one of the foundations of biomedical signal diagnosis in general and especially the PCG signal. In this paper, this signal will be subject of a discrete wavelet decomposition in order to eliminate the unnecessary components contained in the signal, which were determined using energy calculations of wavelet coefficients, After a reconstruction of the signal, we extract the cepstral coefficients which are the Mel frequency cepstral coefficients MFCC, delta MFCC, delta-delta MFCC, the Gammatone cepstral coefficients GTCC, delta GTCC, and delta-delta GTCC coefficients. Following that, several feature sets are going to be produced that can help ensemble learning classifiers recognize PCG signals. The models developed in this way were able to achieve an accuracy of up to 87.76% by a holdout cross-validation.