Smolders Andreas, Lomax Tony, Albertini Francesca
Centre for Proton Therapy, Paul Scherrer Institute, 5232 Villigen, Switzerland.
Department of Physics, ETH Zurich, 8092 Zurich, Switzerland.
Phys Imaging Radiat Oncol. 2025 Apr 23;34:100767. doi: 10.1016/j.phro.2025.100767. eCollection 2025 Apr.
Despite its potential, deformable image registration (DIR) is underutilized clinically, especially in time-sensitive cases, due to a lack of comprehensive metrics for assessing solution quality. Here, we propose a metric of physical plausibility, the bone rigidity error (BRE), that penalizes non-rigid transformations within individual bones, based on the assumption that bones do not deform.
The BRE is calculated by segmenting bones individually and isolating the vectors of a deformable vector field within each bone. A rigid registration is least-square fitted to these vectors, and the BRE is calculated as the average deviation of these vectors from the fitted rigid registration. A lower BRE indicates better rigidity preservation. We evaluated the BRE for 6 DIR algorithms on 32 patients with 137 computed tomography (CT)-to-CT registrations across relevant anatomical sites.
The BRE varied widely between DIR algorithms, up to a factor of 3 on average for inhale-to-exhale thoracic CT registration. Despite large BRE differences between anatomical sites within each algorithm, some algorithms consistently outperformed others. Notably, a low BRE was not correlated with poorer image similarity, and the BRE was only weakly correlated to target registration error. Furthermore, we proposed bone-specific inspection thresholds for patient-specific validation. BRE calculation required less than 5.5 s.
The BRE is an automatic, interpretable, fast, and easy-to-implement metric to assist validation of DIR algorithms, which show widely varying performance. It provides a useful complementary metric for patient-specific validation, especially in time-sensitive applications.
尽管可变形图像配准（DIR）具有潜力，但由于缺乏用于评估解决方案质量的综合指标，其在临床上的应用未得到充分利用，尤其是在对时间敏感的病例中。在此，我们提出一种物理合理性指标，即骨刚度误差（BRE），它基于骨骼不会变形的假设，对单个骨骼内的非刚性变换进行惩罚。
通过单独分割骨骼并分离每个骨骼内可变形向量场的向量来计算BRE。对这些向量进行最小二乘拟合刚性配准，BRE计算为这些向量与拟合刚性配准的平均偏差。较低的BRE表示更好的刚度保持。我们在32例患者的137次计算机断层扫描（CT）到CT配准的相关解剖部位上，对6种DIR算法评估了BRE。
DIR算法之间的BRE差异很大，吸气到呼气的胸部CT配准平均高达3倍。尽管每种算法内不同解剖部位之间的BRE差异很大，但有些算法始终优于其他算法。值得注意的是，低BRE与较差的图像相似性无关，并且BRE与目标配准误差仅呈弱相关。此外，我们提出了针对患者特定验证的骨骼特定检查阈值。BRE计算所需时间不到5.5秒。
BRE是一种自动、可解释、快速且易于实施的指标，有助于验证DIR算法，这些算法表现出广泛不同的性能。它为患者特定验证提供了一个有用的补充指标，尤其是在对时间敏感的应用中。
