FDG is known as a tracer which can be accumulated in the brain, the cardiac muscle and the carcinoma tissue. Recently, FDG-PET became use in cancer diagnosis. The advantageous point of FDG-PET is that it can get a whole body images by one time of measurement, and there are a more few quantities of dose than the other conventional CT equipments. However, there are many doubtful shadow on a FDG-PET image because of the organs whose FDG accumulation are high. By this reason, cancer diagnosis with the FDG-PET is very difficult. In this paper, we propose an computer assisted diagnosis method of whole body cancer with the FDG-PET image. This procedure is as follows. At first, we divide FDG-PET images into some regions on the basis of state of intensity level ups and downs. On this account we calculate Gauss curvature and mean curvature of a FDG-PET image. Second, we extract the positive Gauss curvature and the negative mean curvature as the cancer candidate. The last, we compare the cancer candidate with the diagnosis criteria we defined, and we look for the doubtful cancer regions. This technique is still early stage, but we show the availability of this procedure by comparing with the doctor's diagnosis results.