Wiersma R, Tomarken S, Grelewicz Z, Kang H
The University of Chicago, Chicago, IL.
Northwestern University, IL.
Med Phys. 2012 Jun;39(6Part16):3798. doi: 10.1118/1.4735498.
Both infrared marker and 3D facial surface tracking were performed simultaneously on both a head phantom and healthy volunteers. Accuracies of the two 6D real-time tracking systems were quantitatively assessed both spatially and temporally. The suitability of 3D facial surface tracking for providing real-time head positioning to guide a robotic head motion compensation stage was investigated.
A xyz stage was programmed to move a head phantom to approximately 200 points in 10×10×10 mm and 100×100×100 mm cubic volumes, respectively. Both 3D surface and IR marker tracking were performed simultaneously on the head phantom to allow direct comparison of spatial accuracy. Sinusoidal motion was performed along all three axis directions and recorded under the same clock cycle. Spatial and temporal accuracies on actual deformable facial surfaces were assessed by simultaneous 3D surface and IR marker tracking on human volunteers. 3D surface tracking both with and without the use of robotic head motion compensation was evaluated.
For 10 samples at each position, the 10 mm cube had RMSE of 0.30 mm and 0.08 mm for the AlignRT and Polaris, respectively. Temporal lag of the AlignRT system was found to be approximately twice the fps time (t_{lag} ∼ 2/fps). The fps value was found to be strongly correlated with region-of-interest (ROI) surface size and could vary from 0.5 - 1.6 fps, leading to a lag time of 1 - 4 s. The Polaris output lagged behind the reference by 0.06 s in all cases.
Both 3D surface and IR marker devices demonstrate high spatial accuracy for phantom based studies, however, a slight dimensional warping along the SI direction was observed to occur with the AlignRT. The current ROI dependent lag time of the AlignRT could also pose an issue for image guidance techniques requiring quick response times.
在头部模型和健康志愿者身上同时进行红外标记和三维面部表面跟踪。对这两种六维实时跟踪系统的精度在空间和时间上进行了定量评估。研究了三维面部表面跟踪在为机器人头部运动补偿阶段提供实时头部定位方面的适用性。
对一个xyz平台进行编程，使其分别将头部模型移动到10×10×10毫米和100×100×100毫米立方体内的大约200个点。在头部模型上同时进行三维表面和红外标记跟踪，以便直接比较空间精度。沿所有三个轴方向进行正弦运动，并在同一时钟周期下记录。通过在人类志愿者身上同时进行三维表面和红外标记跟踪，评估实际可变形面部表面的空间和时间精度。评估了使用和不使用机器人头部运动补偿时的三维表面跟踪情况。
在每个位置的10个样本中，对于AlignRT和Polaris，10毫米立方体的均方根误差分别为0.30毫米和0.08毫米。发现AlignRT系统的时间滞后约为帧率时间的两倍（t_{lag} ∼ 2/fps）。发现帧率值与感兴趣区域（ROI）表面大小密切相关，并且可以在0.5 - 1.6 fps之间变化，导致滞后时间为1 - 4秒。在所有情况下，Polaris的输出比参考滞后0.06秒。
三维表面和红外标记设备在基于模型的研究中均显示出较高的空间精度，然而，观察到AlignRT沿SI方向存在轻微的尺寸变形。AlignRT当前依赖于ROI的滞后时间对于需要快速响应时间的图像引导技术也可能构成问题。
