Enhancing Chronic Obstructive Pulmonary Disease (COPD) Diagnosis through Machine Learning Models Trained on Respiratory Sounds

Chronic Obstructive Pulmonary Diseases (COPDs) pose a significant challenge in the medical field due to the complexity of auscultation, which often leads to misdiagnosis by medical professionals. This study aims to address this issue by developing a machine learning-based solution to accurately detect COPD using respiratory sounds. Utilizing a dataset of 920 annotated audio samples collected from 126 patients, we applied various audio feature extraction techniques, including Mel-Spectrograms, MFCCs, Chromagrams, and CENS, to convert the audio classification problem into an image classification task. Several models, including Convolutional Neural Networks (CNNs) and Support Vector Machines (SVMs), were trained on these features. Our results demonstrate that the SVM model trained on augmented MFCC features achieved an accuracy of 89.67%, surpassing the average accuracy of medical professionals. This research highlights the potential of machine learning to enhance diagnostic accuracy in the medical domain, particularly in the detection of COPDs, thereby improving patient outcomes and reducing the burden on healthcare systems.