Heart sound reduction from lung sound recordings applying signal and image processing techniques in time-frequency domain

In this study, two novel HS cancellation methods, using spectrogram Independent Component Analysis (ICA)-based technique and spectrogram filtering-based method in Time-Frequency (TF) domain along with three new techniques for HS localization in respiratory sound recordings are presented. To separate HS from lung sound, the spectrogram ICA-based method applies the ICA algorithm independently to every frequency on the spectrogram of two simultaneously lung sound recordings from two different locations on the chest and yields the independent components at that frequency. Then the proper independent components from each frequency are chosen and combined with each other to produce the spectrogram of separated signals. By implementing Inverse Short Time Fourier Transform (ISTFT), the separated signals are reconstructed in time domain. On the other hand, the spectrogram filtering-based method detects the HS-included segments in the spectrogram of a recorded lung sound signal using one of the proposed HS localization techniques. Afterwards, the algorithm removes those segments and estimates the missing data via a 2D interpolation in the TF domain. Finally, the signal is reconstructed into the time domain. The efficiency of the proposed methods for HS localization and HS cancellation from lung sound recordings was examined quantitatively and qualitatively by visual and auditory means. The results show that the spectrogram ICA-based method is promising in term of HS reduction from lung sound recordings and the spectrogram filtering-based method successfully removes HS from lung sound signals, while preserving the original fundamental components of the lung sounds. The computational cost and the speed of both proposed methods were found to be much more efficient than other HS reduction methods. (Abstract shortened by UMI.).