Lung Sound Recognition Method Based on Wavelet Feature Enhancement and Time-Frequency Synchronous Modeling.

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Signal Processing

Background:

• Accurate lung disease diagnosis is crucial for patient health.
• Artificial intelligence (AI) shows promise in analyzing lung sounds for diagnosis.
• Current AI models often overlook time-domain and frequency-domain correlations in lung sounds and lack detailed feature extraction.

Purpose of the Study:

• To develop an AI model that enhances lung sound analysis by integrating time-domain and frequency-domain information.
• To improve the accuracy and detail of lung sound feature extraction for better disease diagnosis.

Main Methods:

• Proposed a novel AI framework incorporating a dual wavelet analysis module (DWAM) for detailed feature extraction.
• Employed a cubic network with gated recursive units for time-frequency synchronous modeling of lung sounds.
• Integrated an attention module with temporal and channel attention to refine feature representation.

Main Results:

• The proposed model demonstrated superior performance on a combined dataset, outperforming existing methods by over 1.36%.
• On the International Conference on Biomedical and Health Informatics 2017 dataset, the model achieved an improvement of more than 4.28%.

Conclusions:

• The developed AI framework effectively captures complex lung sound features by leveraging wavelet enhancement and time-frequency synchronous modeling.
• This approach offers a significant advancement in AI-driven lung disease diagnosis, surpassing current benchmarks.