A new automatic segmentation method for lung tumor based on SUV threshold on 18 F-FDG PET images

Heart metabolizes more glucose and “light up” more brightly on PET lung images than other healthy or normal tissues and organs because of its constant pumping of blood and accumulation of radiotracer concentration. It affects the results of lung tumor detection and segmentation using the thresholding methods, which is the most used methodcurrently. Standardized uptake value (SUV) on 18 F-FDG PET images isan important indicator to differentiate malignant from benign tumors, and has been applied in tumor volume segmentation. However, some methods based on SUV have been shown unable to distinguish regions between tumor and heart. In this paper, we present a novel segmentation method based on SUV to extract the tumor alone from the background ofhealthy tissues and image noise. Firstly we use SUV to enhance the images and then apply an iterative thresholding method to get a coarse result. Secondly we label the fragmented parts and fuse them with the SUV distribution image, and then calculate the SUV mean for each part. Finally we use the maximum of SUV mean to locate the tumor region. The proposed method is compared with current segmentation methods on PET images. The results show that the new method is able to detect tumors in the background of heart effectively, and can potentially be used as a toolof automatic segmenting tumor from 18 F-FDG PET images.