Phonocardiogram Signal Denoising Using Dictionary Learning

We presented a denoising framework for Phonocardiogram (PCG) signals, a critical component in non-invasive cardiac diagnostics. PCG signals, which capture valuable acoustic information from heartbeats, often suffer from noise contamination due to various physiological and environmental sources. To address this, we applied K-Singular Value Decomposition (K-SVD) and Orthogonal Matching Pursuit (OMP) within a dictionary learning approach to enhance signal quality. Our method leverages sparse representation, iteratively optimizing a dictionary tailored to the underlying clean signal structure. Experimental evaluation on benchmark datasets demonstrates that our framework considerably improves (SNR) and (PSNR), preserving essential diagnostic features of the PCG signals. This approach provides an efficient, clinically valuable solution for improving the clarity and reliability of heart sound analyses, contributing to more accurate cardiovascular diagnoses.