Fractal analysis for quantitative evaluation of diffuse lung abnormalities on chest radiographs: use of sub-ROIs

We evaluated the usefulness of fractal analysis for quantitative evaluation of diffuse lung abnormalities seen on chest radiographs. Regions of interest (ROIs) with 296 x 296 matrix size were selected in the right upper and lower lung zones of digitized chest radiographs. We selected ROIs of 50 mild and 50 severe diffuse lung abnormalities and 50 ROIs of normal lungs. The fractal dimension (FD) for each ROI was defined as the mean value of those obtained from sub-ROIs in each ROI. In the sub-ROIs with 256 x 256, 128 x 128, and 64 x 64 matrix size, the mean values of FDs obtained from the patients with diffuse lung abnormalities were significantly higher than those obtained from the patients with normal lungs (P < 0.001, respectively). The performance of FDs obtained from the sub-ROIs with 256 x 256 matrix size was equal to that obtained from the sub-ROIs with 128 x 128 matrix size (P = 0.980). However, the performance of FDs obtained from the sub-ROIs with 128 x 128 matrix size was superior to that obtained from the sub-ROIs with 64 x 64 matrix size (P = 0.0001). Fractal analysis is useful for quantitative evaluation of diffuse lung abnormalities seen on chest radiographs. From our results, the sub-ROIs whose matrix sizes were greater than or equal to 128 x 128 were suggested to be suitable for quantitative evaluation of diffuse lung abnormalities seen on chest radiographs.