Cellular Automata Segmentation of Brain Tumors on Post Contrast MR Images
A. Hamamcı, G. Unal, K. Engin, N. Kucuk
Segmentation, Brain Tumors, T1 Weighted MRI
A quantitative analysis of brain tumors is an important factor that can have direct impact on a patient’s prognosis and treatment. In order to achieve clinical relevance, reproducibility and especially accuracy of a proposed method have to be tested. We propose a framework for the generation of realistic digital phantoms of brain tumors of known volumes and their incorporation into an MR dataset of a healthy volunteer. Deformations that occur due to tumor growth inside the brain are simulated by means of a biomechanical model. Furthermore, a model for the amount of edema at each voxel is included as well as a simulation of contrast enhancement, which provides us with an additional characterization of the tumor. A “ground truth” is generally not available for brain tumors. Our proposed framework provides a flexible tool to generate representative datasets with known ground truth, which is essential for the validation and comparison of current and new quantitative approaches. Experiments are carried out using a semi-automated volumetry approach for a set of generated tumor datasets.