Bosma Lando S, Ries Mario, Denis de Senneville Baudouin, Raaymakers Bas W, Zachiu Cornel
Department of Radiotherapy, UMC Utrecht, Heidelberglaan 100, 3508 GA Utrecht, The Netherlands.
Imaging Division, UMC Utrecht, Heidelberglaan 100, 3508 GA Utrecht, The Netherlands.
Phys Imaging Radiat Oncol. 2023 Aug 20;27:100483. doi: 10.1016/j.phro.2023.100483. eCollection 2023 Jul.
Deformable image registration (DIR) is a core element of adaptive radiotherapy workflows, integrating daily contour propagation and/or dose accumulation in their design. Propagated contours are usually manually validated and may be edited, thereby locally invalidating the registration result. This means the registration cannot be used for dose accumulation. In this study we proposed and evaluated a novel multi-modal DIR algorithm that incorporated contour information to guide the registration. This integrates operator-validated contours with the estimated deformation vector field and warped dose.
The proposed algorithm consisted of both a normalized gradient field-based data-fidelity term on the images and an optical flow data-fidelity term on the contours. The Helmholtz-Hodge decomposition was incorporated to ensure anatomically plausible deformations. The algorithm was validated for same- and cross-contrast Magnetic Resonance (MR) image registrations, Computed Tomography (CT) registrations, and CT-to-MR registrations for different anatomies, all based on challenging clinical situations. The contour-correspondence, anatomical fidelity, registration error, and dose warping error were evaluated.
The proposed contour-guided algorithm considerably and significantly increased contour overlap, decreasing the mean distance to agreement by a factor of 1.3 to 13.7, compared to the best algorithm without contour-guidance. Importantly, the registration error and dose warping error decreased significantly, by a factor of 1.2 to 2.0.
Our contour-guided algorithm ensured that the deformation vector field and warped quantitative information were consistent with the operator-validated contours. This provides a feasible semi-automatic strategy for spatially correct warping of quantitative information even in difficult and artefacted cases.
可变形图像配准（DIR）是自适应放疗工作流程的核心要素，在其设计中整合了每日轮廓传播和/或剂量累积。传播的轮廓通常需要人工验证并可能进行编辑，从而使配准结果在局部失效。这意味着该配准不能用于剂量累积。在本研究中，我们提出并评估了一种新颖的多模态DIR算法，该算法纳入轮廓信息以指导配准。这将经过操作员验证的轮廓与估计的变形矢量场和扭曲剂量相结合。
所提出的算法包括基于归一化梯度场的图像数据保真项和基于光流的轮廓数据保真项。引入亥姆霍兹 - 霍奇分解以确保解剖学上合理的变形。该算法针对不同解剖结构的同对比度和交叉对比度磁共振（MR）图像配准、计算机断层扫描（CT）配准以及CT到MR配准进行了验证，所有这些均基于具有挑战性的临床情况。评估了轮廓对应性、解剖学保真度、配准误差和剂量扭曲误差。
与没有轮廓引导的最佳算法相比，所提出的轮廓引导算法显著且大幅增加了轮廓重叠，使平均一致距离降低了1.3至13.7倍。重要的是，配准误差和剂量扭曲误差显著降低，降低了1.2至2.0倍。
我们的轮廓引导算法确保了变形矢量场和扭曲的定量信息与经过操作员验证的轮廓一致。这为即使在困难和存在伪影的情况下对定量信息进行空间正确的扭曲提供了一种可行的半自动策略。
