Morphological Analysis of Tumor Regression and Its Impact on Deformable Image Registration for Adaptive Radiotherapy of Lung Cancer Patients

Abstract

Deformational and mass changes associated with regression of the visible tumor during the course of fractionated radiotherapy have confounded the ability to perform accurate deformable image registration, subsequently limiting the clinical implementation of adaptive radiotherapy. This study sought to investigate the impact of tumor regression on the accuracy of deformable image registrations (DIR) and then to find a solution to improve the performance of DIR for treatment of lung cancer patients. Specifically, daily cone-beam computed tomography (CBCT) images were acquired from three locally advanced NSCLC patients. DIRs were performed from fractions 1, 10, and 20 to fraction 25 using a B-Spline-based algorithm implemented within the VelocityAI platform. To improve the accuracy of the BSpline- based registrations in the region of regressing tumors, a hybrid finite element method (FEM) was developed with a mesh defined in a bounding box surrounding the tumor in the target image. The constraints of the FEM model were derived from the displacements generated by the B-Spline registrations. Using the displacement vector fields (DVFs) of the B-Spline and hybrid registrations, the source images were warped to their targets. The accuracies of the two registration algorithms were evaluated, using landmark points identified on both the source and target images, as well as quantitative analysis of the generated DVFs. For the three patients, average tumor volumes were reduced by 53 fraction 1 and fraction 25. Comparison of landmark points showed that the mean errors of the FEM-based hybrid registrations were 1.4, 1.6, and 1.7 mm for the three patients. The average displacement differences between the B-Spline and FEM-hybrid registrations for the three patients were 4.8, 6.2 and 3.9 mm with a maximum of 15 mm for patient 2. Lung tissue does not move consistently with the shrinking tumor. The more the tumor regresses, the larger the B-Spline registration error in the tumor region. The proposed hybrid method that consists of the intensitybased image registration and mechanics-based tissue modeling to correct geometric changes induced by anatomical deformation and tumor regression, respectively, may have the potential to improve the quality of adaptive radiation therapy for lung cancer patients.