2407 - Feasibility of a Laser Scanning System in Reconstructing 3D Surface for Collision Detection during Radiotherapy

Z. Saleh, S. Pani, M. Schmidt, N. Knutson, T. R. Mazur, A. Darafsheh, Y. Hao, T. Zhang, T. Zhao, and S. M. Goddu; Washington University School of Medicine in St. Louis, Department of Radiation Oncology, St. Louis, MO

Purpose/Objective(s): The complexity of treatment delivery has increased in recent years due to the use of non-coplanar beams on linear accelerators (LINACs) and robotic arm couches in proton therapy. Collisions between patient and treatment machine, although rare, may result at minimum in treatment delays and/or disruption of clinical workflow, and at worse, as patient injury. An accurate and comprehensive 3D representation of patient surface and immobilization devices is an essential step towards detecting such collisions. Meanwhile, the CT scans taken during simulation are insufficient in this effort due to limited scan length or truncation of immobilization in the field of view. In this project, we developed a fast and accurate 3D surface scanning system using a laser profiling technique. Materials/

Methods: A bench-top prototype was developed to demonstrate the feasibility of this technique. The system consisted of a red laser with 650nm wavelength and two cameras. Camera-1 is a ViewWorks VH-310G2 (Vision Systems Technology LLC, Vista, CA) camera with a charged-couple device (CCD) sensor and a resolution of 640x480 pixels capable of capturing 264 frames per second (FPS), while camera-2 is a generic HD CMOS-based camera with a resolution of 2592x1944 pixels and 30FPS. The laser was mounted on an elevated rail and pointed down to project a laser line on the object underneath. The cameras were mounted on the same platform at a specified distance from the laser and angled toward the object. A motorized linear rail was constructed to move the object at a fixed speed and the cameras were triggered to capture multiple frames while the object was set in motion. The height profile of the object was calculated by the distortion level of the laser line projection viewed by the camera. The captured video frames were processed offline using an image library Toolbox in the programming environment to extract the 3D surface of the object.
Results: The preliminary results show that the system can reproduce the 3D shape of the scanned object. Table-1 shows the comparison between the physical dimensions and the dimensions obtained from scans of two different objects. 3D scans produced from Camera 2 were more accurate due to its high resolution. The errors were within acceptable limits (10mm) to build a collision detection system.
Conclusion: We demonstrated the feasibility of building a 3D surface scanning system using off-the-shelf components. The system can be scaled to capture 3D surface images of patients in the CT Simulator and during treatment setup for collision detection. The accuracy of the system can be improved further with camera calibration to remove distortions. In addition, using different color lasers and multiple cameras with color filters will improve the image quality and reduce occlusion.

Table-1
			Camera-1 (320x240)			Camera-2 (2592x1944)
		Physical Dimension (mm)	Scan Dimensions (mm)	Error (mm)		Scan Dimension (mm)	Error (mm)
Object 1	Dim-1	31	35	4		32	1
	Dim-2	50	53	3		55	5
	Dim-3	125	125	0		125	0
Object 2	Dim-1	57	67	10		53	4
	Dim-2	57	58	1		58	1
	Dim-3	7.5	10	2.5		9	1.5