Phonocardiogram Event-based Delineation Method Using Continuous Wavelet Transform

Objective: Time frequency-based methods have been employed for event detection in audio signals.

Methods: The spectrogram can be used to extract loudness and energy functions by taking the row sum. The loudness function has been differentiated to obtain the on-set events of the audio. This concept could be replicated in the case of bio-signals such as phonocardiograms to extract on-set and off-set boundaries. A continuous Wavelet Transform-based Time-Frequency algorithm for on-set and off-set event detection has been developed for Phonocardiograms and this method has been tested on the Phonocardiogram (PCG) dataset hosted in the Physionet repository. To validate the simulation, real implementation has also been carried out with real heart sound recordings obtained from a Bluetooth-based stethoscope.

Results: The simulation carried out using PCG sounds from the Physionet database gave an F1-score of 99.11%. Similar results were found using the real heart sound recordings from stemoscope with an F1- score of 99.30%.

Conclusion: Experimental evaluation and Simulations both indicate that a Continuous Wavelet Transform-based time-frequency algorithm can be used to derive on-set and off-set events of the sounds in the PCG signal.