A novel method in COPD diagnosing using respiratory signal generation based on CycleGAN and machine learning

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Respiratory Medicine

Background:

• Chronic Obstructive Pulmonary Disease (COPD) diagnosis often relies on late-stage indicators, impacting patient quality of life and increasing healthcare costs.
• Early and accurate detection of COPD is crucial for effective disease management and improved patient outcomes.
• Distinctive acoustic features in respiratory sounds offer potential for non-invasive diagnostic markers.

Purpose of the Study:

• To develop and evaluate a classification method for diagnosing COPD using unique features extracted from respiratory sounds.
• To compare the diagnostic performance of various algorithms for COPD identification based on acoustic signals.
• To leverage inverse transform techniques and deep learning for enhanced COPD detection.

Main Methods:

• Respiratory sounds were segmented into individual breathing cycles and augmented using the CycleGAN model for data diversity.
• Wavelet families and spectral transformations were employed to analyze characteristic respiratory signals.
• Deep learning models, including VGG16, ResNet50, and InceptionV3, were utilized for the classification task.

Main Results:

• The study demonstrated the effectiveness of the proposed classification method for COPD diagnosis.
• The optimal method combined Wavelet Bior1.3 standardization with the InceptionV3 deep learning model.
• This best-performing approach achieved a highly accurate F1-score of 99.75%.

Conclusions:

• Inverse transform techniques integrated with deep learning models provide highly accurate COPD detection.
• AI-powered analysis of acoustic respiratory features shows feasibility for early COPD diagnosis.
• This approach holds promise for improving early intervention strategies and patient management for COPD.