A noval pulmonary function evaluation method based on ResNet50 + SVR model and cough.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Respiratory Medicine

Background:

• Traditional pulmonary function testing is complex and not suitable for daily patient monitoring.
• There is a need for convenient, accessible methods to evaluate respiratory health.
• Cough sounds present a potential non-invasive parameter for assessing lung function.

Purpose of the Study:

• To develop and validate a deep learning and machine learning model for evaluating pulmonary function parameters using cough sounds.
• To establish a simple, rapid, and location-independent method for respiratory health assessment.
• To correlate cough sound features with standard pulmonary function metrics.

Main Methods:

• Collected 371 cough sound segments from 150 individuals for training and testing.
• Utilized Short-Time Fourier Transform (STFT) to convert cough sounds into spectrograms.
• Employed ResNet50 for feature extraction and Support Vector Regression (SVR) for parameter prediction.

Main Results:

• The ResNet50+SVR model achieved high accuracy in predicting key pulmonary function parameters: FEV1 (R²=0.94), FEV1% (R²=0.84), FEV1/FVC (R²=0.68), FVC (R²=0.92), and FVC% (R²=0.72).
• The hybrid model demonstrated superior performance, especially with limited data.
• Cross-validation confirmed the robustness of the model's performance.

Conclusions:

• The ResNet50+SVR hybrid model offers an effective and rapid method for evaluating pulmonary function through cough sounds.
• This technology enables convenient, remote patient monitoring and early screening of respiratory diseases.
• Potential applications include assessing patient condition, postoperative changes, and detecting infectious respiratory diseases.