AUTOMATIC LUNG TUMOR SEGMENTATION ON PET IMAGES BASED ON RANDOM WALKS AND TUMOR GROWTH MODEL

The segmentation of tumor on PET images is an important step for treatment planning process during the radiotherapy. In this paper, we present an automatic segmentation method on PET images based on the random walks (RW) algorithm. We propose an extension of the random walks framework to integrate a tumor evolution information, which is the predicted tumor region resulting from a model for lung tumor growth and response to radiotherapy. The region of interest (ROI) and labeled seeds are automatically generated. Our approach is compared to the well-known 40% thresholding method, an adaptive thresholding method, a statistical method (FLAB), and a traditional RW algorithm. The good performance of our method has been confirmed on 7 lung tumor patients who are treated with radiotherapy.