Room: AAPM ePoster Library
Purpose: Many proton centers use verification CTs (vCT) to monitor volumetric changes. We have developed a system that registers scout and surface images from video cameras taken at the planning CT (pCT) and vCT. We have extended the system to detect motion by interfacing the video camera with AI computer vision software. The system could objectively monitor intra-fractional motion during treatment. We report on our experience with the registration tools and our preliminary testing of the AI computer vision software.
Methods: We have created a surface image/CT scout image fusion software tool that overlays surface images from pCT and vCT. A pair of reference images of the patient setup is captured during initial simulation for comparison during the vCT. The 2D surface imaging tool provides guidance of the patient setup and is fine-tuned with overlaid scout images, mimicking treatment alignment of bony landmarks, prior to CT scanning. The patient monitoring system can automatically measure the patient’s motion based on pre-trained deep learning algorithms and provide a binary (green/red) signal based upon adjustable tolerances.
Results: The registration tool has improved the vCT registration quality based upon comparisons of vCT/pCT pairs for the same site before and after introduction of the tool. The initial results of the patient tracking system will be presented and show potential for monitoring patient motion objectively.
Conclusion: We have developed an inexpensive x-ray and surface imaging based system for vCTs which could be generalized for photon applications. The use of emerging deep learning and computer vision developments provides the opportunity for surface imaging and monitoring systems. AI potentially permits clinics to create their own training sample for the patient setups and customize the tolerances intra-fractional patient motion across their center.