True positive reduction method for cancer diagnosis based on FDG-PET images

Now we are studying the computer aided diagnosis method of cancer nodule on the FDG-PET images. This system aims that semi-automatic cancer nodule detection. But this system does not reach the practical use because of some problems. The measure factor of these problems is the over estimation of the true positive shade. In this paper, we propose the reduction method of true positive shades on the FDG-PET images using the image registration method. This procedure is as follows. At first, we calculate the ups and downs of the intensity undulations, and we divide the FDG-PET images into some regions to get the preliminary cancer candidates. On this account we calculate Gauss curvature and mean curvature of an FDG-PET image. The second, we calculate the distance between the interesting image and the other normal image with the image registration method. We use SID method as the image registration. The third, we estimate the preliminary cancer candidate on the FDG-PET image as to normal tissue or abnormal tissue by the SID value and we get the true cancer candidates. We show the availability of this method by showing the results of the reduced true positive