Deep learning-based lung sound analysis for intelligent stethoscope

Area of Science:

• Medical technology
• Artificial intelligence
• Respiratory medicine

Background:

• Traditional auscultation has limitations like subjectivity and inability to record sounds.
• Digital stethoscopes enable sound storage and sharing, facilitating telemedicine and education.
• Machine learning, especially deep learning, offers automated lung sound analysis for intelligent diagnostics.

Purpose of the Study:

• To provide a comprehensive overview of deep learning algorithms for lung sound analysis.
• To highlight the role of artificial intelligence (AI) in advancing respiratory diagnostics.
• To present an open-source framework for standardizing deep learning workflows in this field.

Main Methods:

• Review of deep learning algorithms applied to lung sound analysis.
• Focus on converting lung sounds to 2D spectrograms for convolutional neural network (CNN) analysis.
• Examination of task categories, public datasets, denoising techniques, and state-of-the-art methods.

Main Results:

• Deep learning enables end-to-end recognition of respiratory diseases and abnormal lung sounds.
• Identified challenges include device variability, noise sensitivity, and model interpretability.
• An open-source framework is provided to enhance reproducibility and standardization.

Conclusions:

• Deep learning holds significant potential for intelligent stethoscope development and automated respiratory diagnostics.
• Addressing current challenges is crucial for widespread clinical adoption.
• The proposed framework aims to foster collaboration and advance research in AI-driven lung sound analysis.