Estimating respiratory motion from CT images via deformable models and priors

Understanding the movement of tumors during breathing is very important for conformal radiotherapy. Without the knowledge of the tumor movement, it is likely that either insufficient dose is delivered to tumors, or unnecessary dose is received by the surrounding normal tissue, or both. However, respiratory motion is very difficult to study by conventional x-ray CT imaging since object motion causes inconsistent projection views, leading to artifacts in reconstructed images. This dissertation focused on developing methods to build four-dimensional (4D) models of patient's anatomy during breathing, especially in thoracic and upper abdominal region, with currently available X-ray imaging techniques.; We explored methods to estimate respiratory motion from a sequence of conebeam X-ray projection views acquired using a slowly rotating cone-beam CT (CBCT) scanner that was integrated into a Linac system. The slowly rotating CBCT scanners have a large volume coverage and a high temporal sampling rate. In the proposed deformation from orbiting views (DOV) approach, we modeled the motion as a time varying deformation of a static reference volume of the anatomy. We then optimized the parameters of the motion model by maximizing the similarity between the modeled and actual projection views. We used a B-spline based motion model. Challenges of this estimation problem include the limited gantry rotation in one breathing cycle, Compton scatter contamination of the projection views and heavy computation, which will be addressed in the dissertation. We conducted computer simulations and a phantom experiment to test the performance of this approach. Both cases achieved estimation accuracies within voxel resolution. We also explored methods that accelerate the optimization procedure.; We researched the 4DCT imaging methods using multi-slice CT (MSCT) scanners and proposed a method to find the temporal correspondences among the unsorted 4DCT images based on internal anatomical motion. Our method used all the CT slices at each table position to estimate internal motion-based sorting indices. Experiments showed that the internal motion-based sorting greatly reduced tissue mismatch presented in the formed CT volumes using the externally recorded surrogates of breathing motion.