Zaman-frekans bölgesi özniteliklerine dayalı derin sinir ağıyla fonokardiyogram sinyalleri gürültü giderme

Signal processing of phonocardiogram (PCG) has been widely utilized in the preliminary diagnosis of cardiac irregularities. The PCG signals undergo segmentation and classification to identify various heart diseases. However, these signals are frequently affected by diverse types of interference, such as ambient noise, movements from both the stethoscope and the subject, speech, and even lung sounds. As a result, it becomes essential to reduce noise in the signals to achieve the most precise classification of heart abnormalities. We present a denoising strategy employing a deep neural network. In this approach, the Short-Time Fourier Transform (STFT) spectral elements of noisy PCG signals are fed into the convolutional neural network. The deep neural network undergoes training using PCG signals from the Pascal Heart Sounds Challenge database, covering normal and murmur heart sounds under various noise conditions. Experimental results indicate that denoising of phonocardiogram signals using deep neural network based on time-frequency domain features segmentation exhibit superior performance compared to the same methodology without time-frequency features.