Analysis of different methods of automatic image registration

Abstract

In radiotherapy, image registration has become a crucial and primary step for visualization of tumour and normal organs in images which enhanced the accuracy of treatment planning and radiation dose delivery. In this research work, various image registration methods are analyzed and more stress is given on optimization based rigid image registration and deformable image registration. Effect of different interpolators (linear, nearest neighbor, B-spline) and similarity metric (mean square error, mutual information, demons) are evaluated. The available commercial treatment planning system does not provide any quality parameter for image registration. In this study quality of image registration is measured in terms of parameters such as mean square error, normalized cross correlation, structure similarity index and mutual information by different algorithms developed in python. The quality of image registration of the algorithm developed in python for RIR and the commercially available algorithm mare evaluated and compared. Online patient setup correction based on image registration of CBCT and CT images are performed to escalate the radiation dose delivery minimizing the possible radiation dose to normal tissue by defining a small margin from CTV to PTV. But during CBCT also the patient receives an unnecessary dose. There are various parameters like kilovoltage, milliamperes, slice thickness etc but in this study the effect of number of frames is studied on quality of image registration so that appropriate number of frames can be used which would result in relevant quality of registration. Moreover deformable registration was performed for CT and CBCT image datasets to perform dose mapping. Dose mapping will help to determine accurate dose received while using multi-fraction and adaptive radiotherapy treatment. newline