Abstract
Contralateral subtraction (C-sub) is one of a number of computer-aided diagnostic techniques for detecting asymmetric abnormalities in chest radiographs. In this technique, pulmonary nodules in chest radiographs can be enhanced by subtracting a left/right reversed mirror image from the original image. In this report, we propose a C-sub technique that employs novel global and local registration methods. We evaluated the subtraction images obtained by applying the proposed C-sub technique to 107 images with nodules in the JSRT database for which the subtlety levels were judged to be "very subtle", "subtle", and "relatively obvious". First, the quality of rib elimination in the subtraction images was evaluated by a radiologist and assigned a point score (3 = acceptable or better, 2 = poor, and 1 = very poor). In this evaluation, 92.5% of cases were scored as 3. Next, whether or not the nodules were clearly depicted at the specified locations in the subtraction images was evaluated by a radiologist, and the quality of nodule depiction was assigned a point score (3 = clearly depicted, 2 = faintly depicted, and 1 = not depicted). In this evaluation, 73.8% of cases were scored as 3 and 7.5% of cases were scored as 2. The time needed to apply the proposed C-sub technique to the 107 images evaluated was 19.4 seconds per image on average using a personal computer with a 2.2-GHz Intel processor.
