Villa Manuel, Sancho Jaime, Rosa-Olmeda Gonzalo, Chavarrias Miguel, Juarez Eduardo, Sanz Cesar
CITSEM, Universidad Politécnica de Madrid, Madrid, 28031, Spain.
Int J Comput Assist Radiol Surg. 2025 May 23. doi: 10.1007/s11548-025-03416-y.
This study proposes a generalization of markerless patient registration in image-guided neurosurgery based on depth information. The work builds on previous research to evaluate the performance of a range of commercial depth cameras and two different registration algorithms in this context.
A multimodal experimental setup was used, testing five depth cameras in seven configurations. Fiducial registration error (FRE) and target registration error (TRE) metrics were calculated using iterative closest point (ICP) and deep global registration (DGR) algorithms. A phantom head model was used to simulate clinical conditions, with cameras positioned to capture the face and craniotomy regions.
The best-performing cameras, such as the D405 and Zed-M+, achieved TRE values as low as 2.36 ± 0.46 mm and 2.49 ± 0.35 mm, respectively, compared to manual registration that obtains a 1.37 mm error. Cameras equipped with texture projectors or enhanced depth refinement demonstrated improved performance. The proposed methodology effectively characterized the suitability of the camera for the registration tasks.
This study validates an adaptable and reproducible framework to evaluate depth cameras in neurosurgical scenarios, highlighting D405 and Zed-M + as reliable options. Future work will focus on improving depth quality through hardware and algorithmic improvements. The experimental data and the accompanying code were made publicly available to ensure reproducibility.
本研究提出了一种基于深度信息的图像引导神经外科手术中无标记患者配准的通用方法。这项工作建立在先前研究的基础上，以评估一系列商用深度相机在此背景下的性能以及两种不同的配准算法。
使用了一种多模态实验装置，测试了七种配置下的五台深度相机。使用迭代最近点（ICP）和深度全局配准（DGR）算法计算基准配准误差（FRE）和目标配准误差（TRE）指标。使用一个仿真头部模型来模拟临床情况，将相机放置在能捕捉面部和开颅区域的位置。
表现最佳的相机，如D405和Zed-M+，与手动配准误差为1.37毫米相比，分别实现了低至2.36±0.46毫米和2.49±0.35毫米的TRE值。配备纹理投影仪或增强深度细化功能 的相机表现出了更好的性能。所提出的方法有效地刻画了相机对于配准任务的适用性。
本研究验证了一个适用于评估神经外科手术场景中深度相机的可适应且可重复的框架，突出了D405和Zed-M+作为可靠选择。未来的工作将集中于通过硬件和算法改进来提高深度质量。实验数据和相关代码已公开，以确保可重复性。
