Skip to Main Content
Tumor volume delineation plays a critical role in radiation treatment planning and simulation. Inaccurately defined target volume may lead to overdosing of normal structures surrounding the tumor, while potentially underdosing cancerous tissue. Conventional 3-dimensional tumor segmentation methods ignore the temporal information present in dynamic Positron Emission Tomography (PET) images and thus may falsely classify equal intensity voxels with completely different time activity curves (TACs) as belonging to the same tissue. We present a novel approach to tumor volume delineation in dynamic PET based on TAC differences. Principal Component Analysis, nonlinear curve-fitting, region growing, and morphological reconstruction are integrated within a single framework for this purpose. A partially-supervised approach is pursued in order to allow an expert reader to utilize the information available from other imaging modalities routinely used in conjunction with PET. In our scheme, this includes the definition of a tumor encompassing mask and selection of a seed site within the suspected tumor, while further delineation is performed by the algorithm automatically. Performance of the proposed algorithm is compared to five other methods. Simulations and a phantom study show that accurate tumor volume delineation can be achieved.