Denoising of Heart Sound Recordings Standard vs. supervised techniques: is cleaner better for classification?

Area of Science:

• Biomedical Signal Processing
• Artificial Intelligence in Healthcare
• Cardiology

Background:

• Heart sound recordings are frequently degraded by broadband noise, hindering accurate analysis.
• Denoising techniques aim to enhance the signal-to-noise ratio (SNR) of heart sounds.
• The impact of denoising on automatic classification of heart sounds requires further investigation.

Purpose of the Study:

• To evaluate the effect of standard and supervised denoising methods on the automatic classification of healthy versus pathological heart sounds.
• To determine if denoising improves or degrades classification performance under various noise conditions.

Main Methods:

• Comparison of classification performance using heart sound recordings pre-processed with standard denoising techniques.
• Evaluation of classification performance using heart sound recordings pre-processed with supervised denoising techniques.
• Analysis of classification accuracy, sensitivity, and specificity across different noise levels and denoising approaches.

Main Results:

• Supervised denoising significantly improved classification performance, achieving up to a 90% sensitivity gain.
• Standard denoising methods, while increasing SNR, were found to be counterproductive in several noise conditions, leading to decreased classification scores.
• Signal-to-noise ratio alone is not a reliable indicator of improved diagnostic quality after denoising.

Conclusions:

• Conventional denoising techniques can lead to misinterpretation of heart sound diagnoses.
• The application of denoising in heart sound analysis must be approached with caution to ensure diagnostic reliability.
• Cleaner signals do not automatically equate to improved diagnostic accuracy; the focus must be on enhancing diagnostic reliability.