An automatic segmentation method using Bark Spectrogram and loudness index identified S1 and S2 peaks with an accuracy of 96.98% and an F1 measure of 97.09% without requiring threshold setup.
The proposed automatic segmentation method using Bark Spectrogram effectively identifies S1 and S2 heart sounds with high accuracy under noisy conditions without requiring threshold setup.
Effect estimate: F1 measure 97.09%
This paper presents automatic method of segmentation of heart sound using the occurrence of the cardiac rhythmic events. Noisy heart sound is filtered using the 6th order Chebyshev type I low pass filter to remove the redundant noise. Bark Spectrogram is calculated from the cardiac signal by converting spectrogram to the bark scale. The bark spectrogram is smoothened and the loudness index is calculated by averaging the amplitude across all frequency bands. The loudness index is smoothened and differentiated to obtain the event detection function. The smoothened event detection function gives the occurrence of the cardiac events namely the first and the second heart sounds. This method is highly effective in identifying peaks S1 and S2 with the segmentation accuracy of 96.98% giving an F1 measure of 97.09%. This method does not require the setting up of any type of threshold. So it is a highly effective type of segmentation under noisy conditions.
Shervegar et al. (Sun,) conducted a other in Phonocardiogram segmentation. Automatic segmentation method using Bark Spectrogram and loudness index was evaluated on Segmentation accuracy and F1 measure for identifying peaks S1 and S2 (F1 measure 97.09%). An automatic segmentation method using Bark Spectrogram and loudness index identified S1 and S2 peaks with an accuracy of 96.98% and an F1 measure of 97.09% without requiring threshold setup.