The diagnostic results of pneumoconiosis on chest radiographs have great influence on the lives of patients, because their occupational compensations are determined according to the Pneumoconiosis Law based on the ILO classification in Japan. Therefore, we have proposed a computer-aided diagnosis (CAD) scheme for pneumoconiosis on chest radiographs based on the ILO classification.
METHOD AND MATERIALS
Our CAD scheme is as follows. First, regions of interest (ROIs) were selected on digitized chest radiographs. Next, minimum directional difference filtering was performed for enhancement of small rounded opacities and reduction of vascular opacities. Third, the candidates of small rounded opacities of pneumoconiotic changes were selected by use of the degree of their circularity. For quantitative estimation, we employed a physical index called the degree of nodule circularity (DNC), which was a ratio of the rounded opacity area to the area of selected ROI. To evaluate the performance of our CAD scheme, we selected 23 patients with pneumoconiotic changes in the lungs and 12 with normal lungs. 23 patients were classified according to the sizes of small rounded opacities in chest radiographs (type p: 9, type q: 9, type r: 5). In each patient, two ROIs were selected in the right upper and lower lung zones. The DNC of each patient was obtained from the mean value of those obtained from both ROIs.
Abnormal lungs with pneumoconiotic changes were well differentiated from normal lungs by our CAD scheme by use of DNCs (P < 0.05). Also by use of DNCs, there were significant differences between normal and type p pneumoconiotic lungs, type p and type q pneumoconiotic lungs, and type q and type r pneumoconiotic lungs (P < 0.05).
Our CAD scheme could differentiate abnormal lungs with pneumoconiotic changes from normal lungs. The physical index DNC obtained from our CAD scheme was well correlated with the sizes of small rounded opacities in chest radiographs. Our CAD scheme is useful for the classification of pneumoconiosis on chest radiographs based on the ILO classification.