Abstract: A case study was conducted of the slicing of 14 cerebroma PET images, of which interesting parts were divided, with a calculation of their volume features followed. Firstly, all pixels in these images were extracted, and then classified according to the self-organizing neural network, the input node of which was determined as 2×2 through repeated experiments. By marking each element point through the self-organizing neural network, it is easier and more accurate to extract images and to calculate the number of elements within the target zone. With the use of Matlab, it built the threedimension chart of the target zone within the 14 slicings. Finally, the volume of target areas was calculated by using the integral principle. Experimental results show this theory is a plausible one with effective outcome, which can offer assistance in clinical diagnosis.

Key words: self organizing artificial neural network； positron emission tomography（PET)； region-of-interest； integral principle； image segmentation