Blind Source Separation in Biomedical Signals Using Variational Methods

2025-06-23

AI Summary

In clinical auscultation, heart and lung sounds frequently overlap, impeding accurate manual identification and increasing diagnostic error risk. To address this, we propose an unsupervised blind source separation method based on variational autoencoders (VAEs), requiring neither labeled data nor domain-specific priors. Our approach learns a structured latent representation of mixed biomedical signals, enabling interpretable separation of cardiac and pulmonary sounds. A probabilistic decoder performs end-to-end encoding and reconstruction directly from raw recordings acquired via digital stethoscopes. To the best of our knowledge, this is the first work to successfully apply variational inference to heart–lung sound separation. Evaluated on clinically realistic simulated patient recordings, our method achieves clear clustering of heart and lung sounds in the latent space while preserving high-fidelity signal reconstruction—achieving spectral feature reconstruction errors below 3.2%, substantially outperforming conventional blind source separation baselines.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Segmentation
• Search and Optimization: Mixed Discrete/Continuous Search

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Semantics and Knowledge: Methods, algorithms and applications for the development of semantic models, knowledge graphs and other forms of structured data models with machine-interpretable semantics
• Search and Retrieval-Augmented AI: Vertical and domain-specific search

Abstract

This study introduces a novel unsupervised approach for separating overlapping heart and lung sounds using variational autoencoders (VAEs). In clinical settings, these sounds often interfere with each other, making manual separation difficult and error-prone. The proposed model learns to encode mixed signals into a structured latent space and reconstructs the individual components using a probabilistic decoder, all without requiring labeled data or prior knowledge of source characteristics. We apply this method to real recordings obtained from a clinical manikin using a digital stethoscope. Results demonstrate distinct latent clusters corresponding to heart and lung sources, as well as accurate reconstructions that preserve key spectral features of the original signals. The approach offers a robust and interpretable solution for blind source separation and has potential applications in portable diagnostic tools and intelligent stethoscope systems.

Problem

• Separates overlapping heart and lung sounds unsupervised
• Encodes mixed signals into structured latent space
• Enables accurate reconstruction without labeled data

Innovation

• Unsupervised variational autoencoders separate biomedical signals
• Latent space clustering identifies heart and lung sounds
• Probabilistic decoder reconstructs signals without labeled data