An improved random forests approach for interactive lobar segmentation on emphysema detection

Emphysema is one of the most widespread diseases in chronic diseases. Early diagnosis is crucial in slowing down the decline in the lung function of patients. Nowadays, it mainly relies on the pulmonary function test, which suffers from two drawbacks: the pulmonary function test cannot reflect the severity of patients with heterogeneous emphysema accurately and diagnose dyspnea patients. Hence, we propose an approach to analyze emphysema based on computed tomography (CT) images, which can detect the location of emphysema on each lung lobe. For the cases that cannot be automatically segmented, a random forests-based multi-task learning method with granular computing perspective is designed for interactive lobar segmentation. The effectiveness of the proposed emphysema detection method is demonstrated with the CT dataset from 93 patients with chronic obstructive pulmonary diseases. The accuracy of the presented lobar segmentation technique is proved on the CT images that cannot segment lobes. The experimental results show that the proposed interactive lobar segmentation method on locate emphysema about lobes could detect early symptoms of emphysema and reduce \(17.2\%\) of missing diagnosis.