Diagnosis Through Adventitious Sounds Detection and Respiratory Phases Identification Using XGBoost

The study proposes a method to assist physicians who are not respiratory specialists to diagnose specific diseases from lung sounds without advanced medical equipment. The method uses frequency feature images to estimate the presence and progression of disease by performing in parallel the process of comprehensively detecting fine crackles suggestive of disease and the process of identifying the respiratory phases in which the abnormal sounds occur. The validation results of this study showed a detection of fine crackles with an accuracy of 0.90 and an identification accuracy of 0.85 for the respiratory phases. These results are integrated to estimate the progression of the disease. The method provides a visual indication of the location of abnormal sounds, which helps physicians and patients understand and explain their diagnosis.