Deformable image registration (DIRART) of targets and critical organs in prostate CBCT-guided adaptive radiotherapy: geometric and dosimetric evaluation

Objectives To evaluate different deformable image registrations (DIRs) for cone beam computed tomography (CBCT)-based dose calculations. Methods The registration accuracy were evaluated using CIRS phantom and five patients. Deformable planning computed tomography (pCT) and CBCT images registration of prostate cancer cases were conducted using 14 default registration methods and Anatomically Constrained Deformation Algorithm (ANACONDA) registration methods. Quantitative indices of image quality were uniformity, contrast-to-noise ratio, and signal-to-noise ratio for the phantom, and structural similarity index measure, mean absolute error, and peak signal to noise ratio (PSNR) for the patients. Registration accuracy was evaluated using Target Registration Error (TRE) and the Dice Similarity Coefficient (DSC). Results Compared to CBCT, dCT demonstrated considerably improved uniformity 75%, 65%, 77%, contrast-to-noise ratio 214%, 213%, 215%, and signal-to-noise ratio of 108%, 107%, 107% relative to those of CBCT in phantoms with iterative optical flow, B-spline, and ANACONDA registration methods, and improved structural similarity index measures by 14%, 9%, 9%, 11%, 21% and 14%, PSNRs 98%, 87%, 87%, 76%, 94%, and mean absolute errors 98%, 87%, 87%, 76%, 97%, 93% in patients for the Iterative Optical Flow (IOF), B-Spline, Systematic Force Demons (SFD), double force demons (DFD), free form (FF) and ANACONDA registration methods, respectively. Whereas the ANACONDA registration methods improved registration accuracy, within 0.2 cm TRE of prostate landmarks and by more than 90% DSC of critical organs, both IOF and B-Spline algorithms failed to reproduce the geometry for patients. Conclusion Therefore, ANACONDA registration has the potential to contribute to the improvement of adaptive radiotherapy.