%0 Journal Article %A Shan Tan %A Laquan li %A Warren DSouza %A Wei Lu %T Optimal tumor segmentation in PET using multiple thresholds %D 2015 %J Journal of Nuclear Medicine %P 1782-1782 %V 56 %N supplement 3 %X 1782 Objectives The performance of the thresholding method heavily depends on the selected threshold. In this study, we proposed an iterative method to calculate the optimal threshold for PET tumor segmentation using information of the segmentation results of multiple thresholds.Methods Based on a commonly used imaging model we showed for the first time that the problem to calculate the optimal threshold for PET tumor segmentation is itself ill-posed and infinite solutions exist. To calculate the optimal threshold, we extracted an extra constrain, which characterizes how the average intensity of the segmented volume varies with the used threshold. We defined this constrain as the Mean-Optimal-Threshold (MOT) curve. With this constrain, the optimal threshold problem becomes well-posed. We designed an iterative algorithm to calculate the optimal threshold for PET tumor segmentation from multi-thresholds according to the MOT curve. This new method was tested on two clinic patient datasets, one with esophagus cancer (20 patients) and another one with non-Hodgkin’s lymphoma (10 patients). For comparison, thresholding methods with 42% and 50% SUVmax as thresholds, Otsu, Fuzzy C-mean Clustering (FCM), Active Contours (AC), Geodesic Active Contours (GAC), and Graph Cuts (GC) were also tested. The dice similarity index (DSI) was calculated to evaluate the performance.Results For both datasets, the proposed iterative algorithm can converge to the optimal threshold typically in 5 iterations, and the corresponding segmentation using the converged threshold has the best performance (DSI=0.85 for the esophagus cancer dataset, and DSI=0.82 for the lymphoma dataset). By comparison, the GC method (the second best one) has only DSIs of 0.68 and 0.65, respectively.Conclusions The proposed constrain makes the optimal threshold problem well-posed and the iterative algorithm can converge to the optimal threshold quickly.Research Support Shan Tan was supported in part by National Natural Science Foundation of China (NNSFC), under Grant Nos. 60971112 and 61375018, and Fundamental Research Funds for the Central Universities, under Grant No. 2012QN086. Wei Lu was supported in part by the National Institutes of Health (NIH) Grant No. R01 CA172638. %U