Can Heart Sound Denoising be Beneficial in Phonocardiogram Classification Tasksƒ

Area of Science:

• Biomedical Engineering
• Signal Processing
• Medical Informatics

Background:

• Computer-aided diagnosis (CAD) systems aim to enhance disease detection efficiency and reduce subjectivity.
• Heart sound analysis for CAD is challenged by low-amplitude signals and noise from artifacts and physiological sources.
• Effective noise reduction is crucial for improving the diagnostic accuracy of heart sound-based CAD systems.

Purpose of the Study:

• To investigate the impact of four denoising algorithms on heart sound classification performance.
• To determine the optimal application of denoising techniques within CAD systems for phonocardiograph signals.
• To assess the effectiveness of denoising as a preprocessing step for heart sound segmentation.

Main Methods:

• Four distinct denoising algorithms were adapted for phonocardiograph signals.
• Algorithms were evaluated based on their objective impact on heart sound classification accuracy.
• Wiener estimation-based spectral subtraction was specifically tested as a preprocessing step for segmentation.

Main Results:

• Direct application of denoising before classification reduced performance by suppressing murmurs.
• Wiener estimation-based spectral subtraction as a preprocessing step improved segmentation and classification.
• This method achieved 96.0% sensitivity, 74.0% specificity, and 85.0% overall accuracy.

Conclusions:

• Denoising can be detrimental if applied directly before classification, as it may remove diagnostically relevant components like murmurs.
• Integrating Wiener estimation-based spectral subtraction as a preprocessing step enhances heart sound segmentation and subsequent classification.
• This optimized denoising approach significantly improves the performance of heart sound-based CAD systems.