IEEE Trans Biomed Eng. 2024 Sep;71(9):2663-2677. doi: 10.1109/TBME.2024.3386704. Epub 2024 Aug 21.
Intraoperative liver deformation poses a considerable challenge during liver surgery, causing significant errors in image-guided surgical navigation systems. This study addresses a critical non-rigid registration problem in liver surgery: the alignment of intrahepatic vascular trees. The goal is to deform the complete vascular shape extracted from preoperative Computed Tomography (CT) volume, aligning it with sparse vascular contour points obtained from intraoperative ultrasound (iUS) images. Challenges arise due to the intricate nature of slender vascular branches, causing existing methods to struggle with accuracy and vascular self-intersection.
We present a novel non-rigid sparse-dense registration pipeline structured in a coarse-to-fine fashion. In the initial coarse registration stage, we introduce a parametrization deformation graph and a Welsch function-based error metric to enhance convergence and robustness of non-rigid registration. For the fine registration stage, we propose an automatic curvature-based algorithm to detect and eliminate overlapping regions. Subsequently, we generate the complete vascular shape using posterior computation of a Gaussian Process Shape Model.
Experimental results using simulated data demonstrate the accuracy and robustness of our proposed method. Evaluation results on the target registration error of tumors highlight the clinical significance of our method in tumor location computation. Comparative analysis against related methods reveals superior accuracy and competitive efficiency of our approach. Moreover, Ex vivo swine liver experiments and clinical experiments were conducted to evaluate the method's performance.
The experimental results emphasize the accurate and robust performance of our proposed method.
Our proposed non-rigid registration method holds significant application potential in clinical practice.
术中肝脏变形给肝脏手术中的图像引导手术导航系统带来了相当大的挑战。本研究解决了肝脏手术中一个关键的非刚性配准问题：肝内血管树的配准。目标是对从术前计算机断层扫描（CT）容积中提取的完整血管形状进行变形，使其与术中超声（iUS）图像中获得的稀疏血管轮廓点对齐。由于细长血管分支的复杂性质，现有的方法在准确性和血管自交方面存在困难。
我们提出了一种新颖的非刚性稀疏-密集配准流水线，采用粗到精的方式构建。在初始的粗配准阶段，我们引入了参数化变形图和基于 Welsch 函数的误差度量，以增强非刚性配准的收敛性和鲁棒性。对于精细配准阶段，我们提出了一种自动曲率算法来检测和消除重叠区域。随后，我们使用高斯过程形状模型的后计算生成完整的血管形状。
使用模拟数据进行的实验结果表明了我们提出的方法的准确性和鲁棒性。对肿瘤目标配准误差的评估结果突出了我们的方法在肿瘤位置计算中的临床意义。与相关方法的对比分析表明了我们方法的准确性和竞争力。此外，还进行了离体猪肝实验和临床实验来评估该方法的性能。
实验结果强调了我们提出的方法的准确和稳健性能。
我们提出的非刚性配准方法在临床实践中具有重要的应用潜力。
