(Tuesday, 7/16/2019) 7:30 AM - 9:30 AM
Room: 225BCD
Purpose: SBRT relies on extremely accurate delivery of ablative radiation doses to the target. The purpose of this study was to develop an automated, fast, and robust algorithm to enable real-time tracking of fiducial markers to enhance treatment accuracy.
Methods: The algorithm had two parts: template generation using prior information, and online tracking using template matching. A database of over 100,000 CBCT projection images (150 CBCT scans of 30 pancreatic SBRT patients) was analyzed. An offline 3D template was generated through backprojection of filtered CBCT projection images. Then, in each projection image, the forward projection of the 3D template was used to identify the fiducial marker location via template matching. We utilized a breathing-based prediction window to enhance the accuracy and robustness of tracking. We quantified the accuracy of the technique by comparing against results from a previously developed and validated tracking algorithm. We also tested whether data from a single pre-treatment CBCT was sufficient for accurate tracking in all subsequent fractions.
Results: The algorithm identified the markers in 100% of projection images with a mean accuracy of 125 ± 119 μm. Tracking based on prior information resulted in slight decreases to uncertainty metrics (global tracking rate decreased from 93% to 77%, full-area half-maximum of the template match increased from 2.9 mm² to 7.3 mm²), yet the breathing-based prediction prevented any corresponding decreases in tracking accuracy. Separation of the algorithm into offline and online steps achieved significant reductions in tracking time (average 43 ms per image).
Conclusion: We developed a fully automated, fast, and robust fiducial tracking algorithm that can localize marker clusters using prior information. This algorithm could be used for online tracking of tumors, leading to increased accuracy of SBRT delivery.
Funding Support, Disclosures, and Conflict of Interest: Funding: This work was funded in part by the National Institutes of Health under award number K12CA086913, the University of Colorado Cancer Center/ACS IRG #57-001-53 from the American Cancer Society, the Boettcher Foundation, and Varian Medical Systems. Conflict of Interest: Jones and Miften report grants from Varian Medical Systems.
Keywords
Not Applicable / None Entered.
Taxonomy
Not Applicable / None Entered.
Contact Email
