Quantification of interstitial lung abnormalities with chest radiography: comparison of radiographic index and fractal dimension.
RATIONALE AND OBJECTIVES: The authors performed this study to evaluate two physical measures used for quantifying interstitial lung abnormalities on chest radiographs: the normalized radiographic index (RI) and the fractal dimension (FD). MATERIALS AND METHODS: The RI and FD were obtained from 50 regions of interest (ROIs) in lungs with mild interstitial abnormalities, 50 ROIs in lungs with severe interstitial abnormalities, and 50 ROIs in normal lungs. The RI was defined as the normalized percentage area of extracted opacities in selected ROIs. FD was calculated with a box-counting algorithm. To extract linear opacities selectively, the authors processed ROIs with four-directional Laplacian-Gaussian filtering and binarization, linear opacity judgment (LOJ), and linear opacity subtraction (LOS). The usefulness of the physical measures for quantifying interstitial lung abnormalities was evaluated with receiver operating characteristic analysis. RESULTS: In normal and mild abnormality groups, observer performance with the RI was worse with LOJ images (area under the receiver operating characteristic curve [Az] = .812 +/- .042) than with LOS images (Az = .912 +/- .028, P < .05), and performance with FD was better with LOJ images (Az = .867 +/- .037) than with LOS images (Az = .750 +/- .048, P < .05). In the normal and severe abnormality groups, performance with RI and FD was better with LOJ images (RI: Az = .992 +/- .007; FD: Az = .968 +/- .016) than with LOS images (RI: Az = .883 +/- .034, P < .001; FD: Az = .767 +/- .047, P < .0001). CONCLUSION: The RI is considered to reflect the sensitivity in the detection of interstitial lung abnormalities on chest radiographs. Conversely, the FD is considered to reflect specifically the linear opacities processed with LOJ.