Fully Convolutional Hybrid Fusion Network With Heterogeneous Representations for Identification of S1 and S2 From...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiology Signal Processing

Background:

• Heart auscultation is a crucial, cost-effective method for initial heart abnormality screening.
• Phonocardiogram (PCG) analysis using computerized algorithms can enhance diagnostic accuracy.
• Accurate identification of S1 and S2 heart sounds is fundamental for diagnosing cardiac conditions.

Purpose of the Study:

• To develop a fully convolutional hybrid fusion network for precise S1 and S2 sound localization in PCG signals.
• To enable high-level, timewise fusion of heterogeneous features (1D envelope and 2D spectral) without temporal distortion.
• To advance the clinical utility of phonocardiogram analysis through improved signal interpretation.

Main Methods:

• Development of a fully convolutional hybrid fusion network.
• Implementation of a novel convolutional multimodal factorized bilinear pooling for feature fusion.
• Fusion of dimensionally heterogeneous features: 1D envelope and 2D spectral data.

Main Results:

• The proposed method successfully identifies S1 and S2 locations in phonocardiograms.
• Demonstrated superior performance compared to existing state-of-the-art PCG segmentation techniques.
• Validated the benefits of comprehensive interpretation via high-level fusion of heterogeneous features.

Conclusions:

• The developed hybrid fusion network offers a significant advancement in phonocardiogram analysis.
• This approach represents a novel method for interpreting heterogeneous features in PCG data.
• The findings support the potential of AI-driven auscultation for early heart abnormality detection.