Deep learning fusion framework for automated coronary artery disease detection using raw heart sound signals

Area of Science:

• Cardiology
• Biomedical Engineering
• Artificial Intelligence in Medicine

Background:

• Coronary artery disease (CAD) is a prevalent cardiovascular condition requiring early diagnosis.
• Current computer-aided detection methods for CAD using heart sounds often depend on subjective expert analysis of graphical representations.
• There is a need for objective and accurate noninvasive methods for early CAD detection.

Purpose of the Study:

• To develop and validate a novel fusion framework for noninvasive coronary artery disease detection using raw heart sound signals.
• To overcome the limitations of subjective expert-based analysis in current CAD detection methods.
• To integrate multidomain and medical multidomain features through deep learning for improved CAD diagnosis.

Main Methods:

• A fusion framework combining two CAD detection models was constructed: a multidomain feature model and a medical multidomain feature fusion model.
• Heart sound signals from 400 participants were collected.
• The framework extracted 206 multidomain features and 126 medical multidomain features, fusing them with one-dimensional deep learning features.

Main Results:

• The multidomain feature model achieved an Area Under the Curve (AUC) of 94.7%.
• The medical multidomain feature fusion model achieved an AUC of 92.7%.
• The fusion framework demonstrated effectiveness in integrating diverse heart sound features via deep learning.

Conclusions:

• The proposed fusion framework offers an effective, noninvasive solution for early coronary artery disease detection.
• Deep learning algorithms can successfully integrate one-dimensional and cross-domain heart sound features for improved diagnostic accuracy.
• This approach reduces reliance on expert subjectivity in CAD diagnosis from heart sounds.