Segmentation of small pulmonary nodules on thoracic CT images

This paper presents a segmentation algorithm of small pulmonary nodules (SPNs) on thoracic CT images for follow-up and differential diagnosis. The proposed method consists of the following two steps: a rough segmentation of an SPN by use of both a variable threshold and a template-matching method, a fine segmentation of the SPN by use of both a watershed method and a distance-transformation method. We quantitatively evaluated the performance of our algorithm using a coincidence rate (the region obtained from computerized segmentation/the region decided by the radiologist). (author)