Feasibility evaluation of radiotherapy positioning system guided by augmented reality and point cloud registration

Abstract

Purpose: To develop a radiotherapy positioning system based on Point Cloud Registration (PCR) and Augmented Reality (AR), and to verify its feasibility.

Methods: The optimal steps of PCR were investigated, and virtual positioning experiments were designed to evaluate its accuracy and speed. AR was implemented by Unity 3D and Vuforia for initial position correction, and PCR for precision registration, to achieve the proposed radiotherapy positioning system. Feasibility of the proposed system was evaluated through phantom positioning tests as well as real human positioning tests. Real human tests involved breath-holding positioning and free-breathing positioning tests. Evaluation metrics included 6 Degree of Freedom (DOF) deviations and Distance (D) errors. Additionally, the interaction between CBCT and the proposed system was envisaged through CBCT and optical cross-source PCR.

Results: Point-to-plane iterative Closest Point (ICP), statistical filtering, uniform down-sampling, and optimal sampling ratio were determined for PCR procedure. In virtual positioning tests, a single registration took only 0.111 s, and the average D error for 15 patients was 0.015 ± 0.029 mm., Errors of phantom tests were at the sub-millimeter level, with an average D error of 0.6 ± 0.2 mm. In the real human positioning tests, the average accuracy of breath-holding positioning was still at the sub-millimeter level, where the errors of X, Y and Z axes were 0.59 ± 0.12 mm, 0.54 ± 0.12 mm, and 0.52 ± 0.09 mm, and the average D error was 0.96 ± 0.22 mm. In the free-breathing positioning, the average errors in X, Y, and Z axes were still less than 1 mm. Although the mean D error was 1.93 ± 0.36 mm, it still falls within a clinically acceptable error margin.

Conclusion: The AR and PCR-guided radiotherapy positioning system enables markerless, radiation-free and high-accuracy positioning, which is feasible in real-world scenarios.

Keywords: ICP; augmented reality; point cloud registration; radiotherapy positioning; visualization.

FIGURE 1

The proposed PCR procedure flow chart.

FIGURE 2

Preprocessing of the proposed PCR procedure. a: Original phantom, b: Point cloud after uniform down‐sampling, c: Statistical filter processing: red points are the discrete point identified by the filter, d: Discrete points elimination.

FIGURE 3

Point‐to‐Plane ICP registration based on FPFH. a: Point cloud after pre‐processing, b: Normal estimation of point clouds: the purple lines are normals, c: FPFH description: red points are the feature points, d: The two‐point clouds before ICP registration: the red point cloud is the target position, and the green is the floating position, e: Two‐point cloud with coincident positions after ICP registration.

FIGURE 4

Phantoms experiment process of AR‐and‐PCR‐guided positioning system simulated radiotherapy positioning.

FIGURE 5

Phantom experiment scene of simulated radiotherapy positioning. a: Location of components on the experimental site, b: CR‐Scan‐Ferret scanning range, c: 3D scanning imaging, d: High‐precision operating platform (Simulated treatment couch).

FIGURE 6

Initial position correction guided by AR and 6DOF deviations.

FIGURE 7

Real human experiment scene of radiotherapy positioning. a: Location of components on the experimental site, b: Accelerator display screen(the each movement of the treatment couch can be read by three values in the yellow dashed box).

FIGURE 8

Interaction envisage of the phantom‐based CBCT with the proposed system. a: CBCT imaging scene of the phantom, b: CBCT image of the phantom, c: ROI extraction of surface point cloud of CBCT, d: Surface point cloud of optical scanning, e: CBCT‐optical cross‐source PCR effect.

FIGURE 9

Line plots of point cloud reconstruction evaluation from 15 patients.