A computerized analysis system in chest radiography: evaluation of interstitial lung abnormalities

We evaluated the usefulness of a computerized analysis system in the detection of interstitial lung abnormalities in digitized chest radiography. This system uses the processes of four-directional Laplacian-Gaussian filtering, linear opacity judgment, and linear opacity subtraction. For qualitative analysis, we employed a combined radiographic index, which was calculated from two normalized radiographic indices obtained by linear opacity judgment and subtraction of linear opacities. We selected 50 regions of interest (ROIs) in patients with mild interstitial lung abnormalities, 50 ROIs in patients with severe interstitial lung abnormalities, and 50 ROIs in individuals with normal lung parenchyma. High-resolution computed radiography (HRCT) findings were used as the standard of reference for this study. These ROIs were processed by our computerized analysis system, and radiographic indices were obtained from each ROI. The area under the receiver operating characteristic curve (Az) was used as the measure of performance. The combined radiographic index provided better results in the mild interstitial lung abnormality group (Az = 0.94 +/- 0.02), but it also yielded good results in the severe interstitial lung abnormality group (Az = 0.98 +/- 0.01). These results indicate that this system of combining radiographic indices has improved the detection performance over that with our previous system.