233. Radiotherapy treatment planning of the pelvis region using only MR pseudo-CT images

Abstract

Introduction Radiotherapy treatment plannings are based on CT images that provide information of tissue density. A fusion between CT and MR images is often performed in anatomical regions characterized by soft tissue prevalence to improve contouring, considering the high contrast provided by MRI. In this study, a method for the reconstruction of pseudo-CT of pelvis region, using only T1/T2-weighted MR images, was developed to reduce radiation exposure, improve anatomical contrast and skip image fusion during planning. Materials and Methods CT and MR images of 6 patients were acquired using a GE LightSpeed Xtra and a GE Optima MR360 1.5 T Advance, respectively. A registration between MR and CT images was performed by SPM12. The images of 5 patients were segmented in 5 different regions, according to the histogram of the CT images: body surface, soft tissue, bone surface, spongy bone and compact bone. Fitting parameters were obtained by the least squares method, comparing MR and CT intensities. Akaike information criterion highlighted the polynomial function that best fitted the data for each segmented region. The evaluated fitting parameters were used to generate pseudo-CT images from the MR images of the 6th patient. Body surface and soft tissue were automatically segmented, while bone structures were identified by a semiautomatic method. For validation, corresponding CT and pseudo-CT voxel intensities were compared. Dose distribution accuracy in a direct field treatment was estimated evaluating the dose in 10 points and PDDs on both CT and pseudo-CT images by Pinnacle3. Results A mean relative intensity difference of ≈1.3% was found between CT and pseudo-CT images, mainly due to body surface and bone segmentation. The evaluated dose and PDD mean relative differences were of ≈1.2% and ≈1%, respectively. Conclusions Our preliminary results on pseudo-CT planning show the method feasibility. These could be improved implementing an automatic bone segmentation.