A Free Form Deformation-based 3D Registration Algorithm for Image Guided Breast Radiotherapy

Abstract

Listen

Translate article

Accelerated partial breast irradiation (APBI) provides an alternative treatment strategy for early stage breast cancer patients. Image guided radiotherapy (IGRT) has the potential to improve treatment accuracy and quality. However, this potential cannot be fully utilized unless the clinical target volume can be quickly identified and localized at treatment. The purpose of this study is to develop a novel and fast registration method based on discriminating and registering the breast soft tissue vs. the seroma cavity, to capture the clinical target volume on the treatment 3D cone-beam CT (CBCT) images from the planning 3D CT images during the partial breast treatment in IGRT. We present a global to local 3D shape registration framework between the planning 3D CT images and treatment 3D CBCT images. The global registration is based on maximizing mutual information and the local registration is based a B-spline based incremental free form deformation model to minimize a sum of squared differences measure. The strengths of this method are that it preserves shape topology during local deformation, develops a novel 3D distance map algorithm and produces smooth, continuous and one-to-one correspondence local registration fields. The method is applied on 7 datasets consisting of both planning CT and CBCT images of breast patients. The target volumes which are used in the registration method in the planning 3D CT images and treatment 3D CBCT images are delineated by radiation oncologists manually and are chosen as the gold standards. The registration method is evaluated using both qualitative and quantitative methods. The qualitative method is based on visualization by radiation oncologists. The quantitative methods are based on distance-based estimators and volume overlap ratio to compare the differences between the gold standard of the treatment CBCT images and the registered planning images. The quantitative results show that between the gold standard of the treatment CBCT images and the registered planning images, the mean discrepancy in distance ranges from 1.21 mm to 2.32 mm, the root-mean-square error (RMSE) ranges from 1.35 mm to 2.54 mm and the mean breast volume overlap ratio ranges from 83.2% to 94%. We present a novel and fast deformable registration method to capture the transformation between the planning and treatment images for external beam radiotherapy. We also present promising results of our method applied to clinical datasets. These preliminary results show that the proposed method is robust and reasonably fast for the registration of the deformable soft tissue of breast, and for deriving the clinical target volume on the treatment 3D CBCT images from the planning 3D CT images during the partial breast treatment.