A Hybrid Deep Learning Approach for Early Detection of Chronic Obstructive Pulmonary Disease

Chronic obstructive pulmonary disease (COPD) is currently the third leading cause of death worldwide. Early detection can help treat the disease and delay its progression. However, chronic diseases are difficult to detect and symptoms often have to develop into severe conditions before they become apparent. Currently, physicians use artificial auscultation as a preliminary means of diagnosing COPD. By detecting the respiratory acoustic phenomena and analyzing the pathology knowledge of patients, physicians can infer and analyze the disease. However, this method still has the possibility of misjudgment or delayed treatment. Therefore, this study uses the ICBHI Respiratory Sound Database Dataset as the basis for analysis data set under the deep learning technology with convolutional neural network models, we classify the features of lung sounds and hope to construct an identification tool that assists in diagnosing COPD. In addition to reducing the time cost of traditional auscultation with this auxiliary tool, after evaluating the model’s effectiveness with confusion matrix and accuracy evaluation, we especially estimate its correctness and practical applicability. In the future, it can be recommended for clinical diagnosis and development of an auxiliary diagnosis tool that helps provide early diagnosis of COPD.