Prediction of tumor deformation for Image Guided Radiation Therapy
Hoog Antink, Christoph Bernhard
MetadataShow full item record
Radiation therapy uses ionizing radiation to damage the DNA of cancerous cells. It is extensively used for curative and palliative treatment and constitutes more than 50% of all lung cancer treatment modalities. Precise knowledge of the shape and location of the tumor is mandatory for effective Intensity Modulated Radiation Therapy. The task of accurately targeting the tumor is especially challenging when the tumor's location and shape is influenced by respiratory motion (e.g. tumors on the lung, breast, etc.). In this work, we develop an image-processing technique that assists the radiation therapy planning process, by exploiting computed tomography images to estimate the shape and position of the tumor over the breathing cycle. To validate the proposed algorithm for Deformable Image Registration, benchmark datasets provided to the community by a research group at The University of Texas M. D. Anderson Cancer Center (http://www.dir-lab.com/) were used. Numerical evaluation illustrates that the proposed algorithm outperforms all algorithms tested on the benchmark dataset for a majority of the cases. To advance the field of on-line tumor tracking from rigid motion towards deformation, time-varying Fourier Descriptors were used to learn a deformation model. Combining these two techniques we further developed a Semi-Automated Contouring software and evaluated it with real patient data provided by the Roswell Park Cancer Institute .