An Improved Contralateral Subtraction Scheme for Detection of Pulmonary Nodules in Chest Radiographs

摘要

Contralateral subtraction (C-sub) is a computer-aided diagnosis technique for detecting pulmonary nodules in chest radiographs. This technique enhances nodules in a chest image by subtracting its right/left reversed mirror image from the original image. In this paper we propose an improved C-sub scheme, which uses a ribcage boundary detection method and a global matching method newly proposed in this paper. First, we evaluated the ribcage boundaries detected by the proposed method using 247 images in JSRT database which is a public database. As the result, the complete ribcage boundaries including the top lung parts were detected in good accuracy for 97.6% images and in fair accuracy for 1.6% images. Next, we evaluated the subtraction images obtained by the proposed C-sub scheme using 107 images with nodules in JSRT database, whose subtlety levels are 'very subtle', 'subtle', and 'relatively obvious'. For the evaluation, a radiologist checked whether or not nodules were clearly depicted in the subtraction images. As the result, the radiologist decided that nodules were clearly depicted for 75.7% images.