Lung tumor delineation in PET-CT images based on a new segmentation energy

Accurate lung tumor delineation from positron emission tomography (PET) - computed tomography (CT) images is important for patient management. However, this task is challenging, especially when a tumor abuts or involves the chest wall or mediastinum; or when it is located in close proximity to the heart or liver. Combining PET and CT can improve the accuracy of tumor delineation, but current methods might be prone to leakage or require that the tumor already be isolated within a box. In this paper, we proposed a tumor delineation method with a novel energy function using information from both CT and PET with a downhill cost. We added a penalty to the energy function such that when a voxel is not part of the downhill region, the energy will be the negative of the initial energy. With this formulation, the contour did not move outside the downhill region thus delineation leakage can be avoided. We incorporated our proposed energy function into a region-based active contour to address the issue of insufficient gradient information when a tumor is connected to surrounding organs of similar intensity. We implemented the segmentation using a localized active contour to focus on the voxels near the tumor, without considering all the global intensities. Out method only required a seed point from which the maximum intensity in PET can be found and used to define a downhill region for the energy function. We evaluated our method in 20 PET-CT studies, from which 10 studies with manual delineation were used for quantitative evaluation and 10 studies where the tumors abutted or involved the chest wall, mediastinum or diaphragm were used for qualitative evaluation. Experiments demonstrated that our method: (1) achieved the highest overlap compared to other approaches that used PET only, CT only, and combined PET-CT without the downhilAccurate lung tumor delineation from positron emission tomography (PET) - computed tomography (CT) images is important for patient management. However, this task is challenging, especially when a tumor abuts or involves the chest wall or mediastinum; or when it is located in close proximity to the heart or liver. Combining PET and CT can improve the accuracy of tumor delineation, but current methods might be prone to leakage or require that the tumor already be isolated within a box. In this paper, we proposed a tumor delineation method with a novel energy function using information from both CT and PET with a downhill cost. We added a penalty to the energy function such that when a voxel is not part of the downhill region, the energy will be the negative of the initial energy. With this formulation, the contour did not move outside the downhill region thus delineation leakage can be avoided. We incorporated our proposed energy function into a region-based active contour to address the issue of insufficient gradient information when a tumor is connected to surrounding organs of similar intensity. We implemented the segmentation using a localized active contour to focus on the voxels near the tumor, without considering all the global intensities. Out method only required a seed point from which the maximum intensity in PET can be found and used to define a downhill region for the energy function. We evaluated our method in 20 PET-CT studies, from which 10 studies with manual delineation were used for quantitative evaluation and 10 studies where the tumors abutted or involved the chest wall, mediastinum or diaphragm were used for qualitative evaluation. Experiments demonstrated that our method: (1) achieved the highest overlap compared to other approaches that used PET only, CT only, and combined PET-CT without the downhill cost; and (2) avoided leakage in all cases when other approaches suffered leakage.l cost; and (2) avoided leakage in all cases when other approaches suffered leakage.