Research On Medical Image Registration Based On4D-CT And Clinical Application In Radiation-induced Liver Disease

Radiation therapy acts an important role for hepatocellular carcinoma patients, with the development of radiotherapy, precise radiotherapy can increase the dose of tumor targets on the basis on decreasing the dose of normal liver and organs of risk, how to predict the probability of radiation-induced liver disease (RILD) effectively has arose lots of research. The common-used currently dose-volume indices to predict probability of RILD was calculated by treatment planning system based on static computed tomography scanning data. The indices contained some uncertainties because the static computed tomography could not reflect the breath motion and the organ deformation accurately, and the uncertainties could affect the accuracy in predicting RILD. Recently four dimensional computed tomography (4D-CT) is dynamic CT images which related to time could truly reflect the motion trajectory of the liver and target volume from breath motion. The changes of position, shape and volume of normal liver from breath motion must be worked out during dose accumulation, and deformation registration is an effective approach.This paper proposed two registration methods:FFD model registration method based on B-Spline and diffeomorphic deformable registration method based on demons.(1) FFD model registration method based on B-Spline first used affine transformation to accomplish the global registration, FFD model registration method based on cubic B-Spline registered the different phases images of4D-CT of HCC patients. The B-spline has an advantage of local controlling, and FFD has good smooth restraining, robustness to noise and one to one mapping. Considering the huge amount of computation, we used limited-memory L-BFGS method to optimize the system, and multi-resolution was used to improve the registration accuracy and robustness.(2) The original demons algorithm is a small deformation model which cannot handle large deformation, and it breaks the topology of original images to affect anatomy analysis during the registration process. Diffeomorphic algorithm was combined to work out these to achieve the high computing efficiency and satisfied registration performance. Furthermore, the recent methods always lack landmarks to verify the registration performance, and the majority of HCC patients are received transcatheter arterial chemoembolization (TACE). So we used iodipin to improve the registration method and verify the performance of registration method as the markers.Two registration methods were applied to register the different phases images of4D-CT, which supplies the theoretic foundation for dose accumulation point to point, the prediction could be more accurate and objective, and the anatomic position and the range of RILD could be predicted precisely.