Room: AAPM ePoster Library
Purpose: To determine if radiomics analysis of CBCT images of a Multiple Imaging Modality Isocentricity (MIMI) phantom (Standard Imaging) acquired for daily quality assurance (QA) geometric tests can be used to retrospectively evaluate kV x-ray tube performance.
Methods: CBCT projections acquired using head scan protocols on a TrueBeam LINAC were reconstructed using iTools Reconstruction (Varian Medical Systems). 326 scans spanned a two-year date range (12/1/16-11/30/18). The central slice of each scan was identified and a 222-by-222 pixel region-of-interest, encompassing approximately 60% of the central area, was automatically selected for feature extraction. PyRadiomics was used to extract features such as: gray level co-occurrence matrix (GLCM) sum entropy, first-order uniformity, and first-order mean.
Results: All three aforementioned features plotted over time exhibited two distinct shifts from baseline. The first shift occurred around July 3, 2017 and the second on July 2, 2018. The first shift corresponded to kV image quality degrading and routine monthly CBCT QA intermittently failing in three nonconsecutive failures in a 12-month span. Further, GLCM sum entropy and first-order mean both decreased by 0.74%±0.38% and 1.49%±0.86%, respectively, while first-order uniformity had the largest variation, increasing by 2.28%±0.98%. The monthly QA data over the same time period lacked the sensitivity or sampling frequency to detect this change. The second shift corresponded to tube replacement.
Conclusion: Analysis of radiomics features indicates that kV x-ray tube replacement directly impacted select components of image texture in daily but not monthly QA data. Radiomics analysis of phantom QA images acquired at high frequency (i.e., daily) over an extended period of time could be used in conjunction with conventional quality metrics to track the overall performance of CBCT equipment.
Funding Support, Disclosures, and Conflict of Interest: This work was supported in part by the University of Chicago BSD Summer Fellowship, the NIH T-32 Training Grant T32 EB002103, and Varian Medical System. Further, HA receives royalties and licensing fees for computer-aided diagnosis technology through the University of Chicago.