P Brown*, S Tanny, D Rosenzweig, University of Rochester, Rochester, NY

• 
  P Brown

PO-GePV-T-183 (Sunday, 7/25/2021)   [Eastern Time (GMT-4)]

Purpose: To develop a large dataset that explores the parameter space of gamma criterion using our institutional patient-specific quality assurance (PS-QA) data from the past five years of using EPID-based QA. This dataset is to be used to train and test a machine learning model that will be used to identify plans that may require additional quality review.

Methods: A Portal Dosimetry Application Programming Interface (PDAPI, Varian Medical Systems, Palo Alto, CA, USA) application has been developed to process and extract PS-QA results applying various gamma evaluation criteria. All 27 combinations of gamma criteria involving dose differences of 1, 2, and 3%, distance-to-agreement of 1, 2, and 3 mm, and evaluation thresholds of 10, 20, and 50% of dose maximum were applied to EPID-QA images acquired at our institution since 2015. We collected average, maximum, percentage of points with gamma >1, and >1.2 gamma values for all image-criteria combinations. Average and variance value for each QA session were also calculated for comparison with values presented in the literature.

Results: Our dataset consists of 94000 analyses with >3300 images involving >700 patients. Average percentage of points with gamma >1 for 3%/3mm/10%threshold was 3.8%, which agrees with published results of 4.1% in smaller series from other institutions. The higher the threshold for allowable points, the worse the gamma results were for all evaluations. Trending was performed demonstrating an improvement in gamma results over time from 202/735 images to 100/581 images with area of gamma >1 <90% using TG-218 criteria of 3%/2mm/20% Threshold.

Conclusion: We have automated a gamma analysis data extraction tool and will continue to build our dataset to use in future studies evaluating contributing factors for poor gamma passing rates using EPID-based QA. Our current dataset is amongst the largest extracted datasets of patient specific QA results in literature.

Keywords

Not Applicable / None Entered.

Taxonomy

Not Applicable / None Entered.

Contact Email