Mutual Information Registration Between Portal And Reference Images In Radiotherapy

The purpose of the external radiation in radiotherapy is to kill as many tumor cells as possible with the least damage to normal tissues. However, the inappropriate setting of the patient in the course of therapy results in the radiation error between the target position and the practical one. This error lowers radiotherapy quality because target tumor is not fully exposed to radiation. Furthermore, the inappropriate radiation may cause normal tissues damages. Radiotherapy quality can be improved by rapid assessment of the position and timely correction of set-up error. Usually the set-up error of radiotherapy is analyzed with image registration, i.e., registration of the practical portal image with the prospective simulated image or planning images. Comparing several kinds of registration method, this paper proposed the mutual information registration method based on features to register the portal images and reference images, according to the analysis of setting errors in radiation therapy.Edge features extraction is the key technique in the mutual information registration method based on features. In this paper, Canny Operator with preferable anti noise and edge position was used to extract edge features. The extracted edge features were set as datum mark to calculate the maximal mutual information between the portal images and reference images. Parameters were optimized with simplex-simulated annealing optimization strategy. We have also done some improvements to the simplex-simulated annealing optimization strategy to meet the registration speed requirement.We tested the feasibility of the registration algorithm, using the simulatedimages with various levels of SNR from -1.5168dB to 15.693dB. The results showed that the registration precision in this paper reached sub-pixels level with the extension of SNR above OdB.In this paper, to improve image effects, the image was processed by the gray-scale transformation combined with median filter before registration. The mutual information registration method based on features was applied on the clinical registration for radiation portal images and reference images. This paper compared the results of the mutual information based on features with the results of the same method based on gray-scale. Although the registration precision of the mutual information registration method was little lower than that of the method based on gray-scale, the registration speed was increased remarkably. So this mutual information registration method based on features can be adapted for online registration of clinical radiation images.