Rapid detection and interpretation of heart murmurs using phonocardiograms, transfer learning and explainable...

Area of Science:

• Cardiology
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Cardiovascular disease is a leading cause of mortality.
• Early diagnosis of heart sounds, including murmurs, is crucial for prevention.
• Computer-aided systems can enhance murmur detection and decision-making.

Purpose of the Study:

• To develop and evaluate an AI model for detecting and classifying heart murmurs from phonocardiograms.
• To assess the model's ability to predict murmur presence, absence, and severity.
• To interpret the AI model's decisions using Explainable Artificial Intelligence (XAI).

Main Methods:

• Phonocardiogram signals were converted into Mel spectrograms.
• The OpenL3 network was used for transfer learning to classify the spectrograms.
• Murmur severity was classified using pitch level and Levine scale.
• Occlusion Sensitivity was employed as an XAI method for model interpretation.

Main Results:

• The AI model achieved impressive results in classifying murmurs without prior signal segmentation.
• The model successfully predicted the presence/absence and severity levels of murmurs.
• XAI analysis revealed the specific features the neural network utilized for decision-making.

Conclusions:

• AI, particularly using Mel spectrograms and transfer learning, shows significant potential for automated heart sound analysis.
• XAI methods are essential for understanding and trusting AI-driven diagnostic tools in cardiology.
• This approach can provide detailed insights into phonocardiogram features for improved cardiovascular diagnostics.