Co-Segmentation Methods for Improving Tumor Target Delineation in PET-CT Images

Abstract

Positron emission tomography (PET)-Computed tomography (CT) plays an important role in cancer management. As a multi-modal imaging technique it provides both functional and anatomical information of tumor spread. Such information improves cancer treatment in many ways. One important usage of PET-CT in cancer treatment is to facilitate radiotherapy planning, for the information it provides helps radiation oncologists to better target the tumor region. However, currently most tumor delineations in radiotherapy planning are performed by manual segmentation, which consumes a lot of time and work. Most computer-aided algorithms need a knowledgeable user to locate roughly the tumor area as a starting point. This is because, in PET-CT imaging, some tissues like heart and kidney may also exhibit a high level of activity similar to that of a tumor region. In order to address this issue, a novel co-segmentation method is proposed in this work to enhance the accuracy of tumor segmentation using PET-CT, and a localization algorithm is developed to differentiate and segment tumor regions from normal regions. On a combined dataset containing 29 patients with lung tumor, the combined method shows good segmentation results as well as good tumor recognition rate.