Cherenkov Imaging to Verify Radiation Treatment Delivery in Breast Cancer

Abstract

<h3>Purpose/Objective(s)</h3> Cherenkov imaging can record beam shapes on patients, assessing patient positioning and dose accuracy. Previous research conducted has compared each treatment fraction to the first fraction, to confirm consistency over the course of treatment. New advancements in Cherenkov imaging now allows for comparing fractions to the planning system's predicted surface dose. We compared predicted surface dose to Cherenkov images to assess for treatment accuracy quantitatively and qualitatively, and to test this new methodology. <h3>Materials/Methods</h3> All data was obtained with IRB approval from patients receiving breast radiation therapy with 6X and/or 10X beams. The predicted surface dose was obtained by exporting the simulation CT scan and field-specific dose distributions from the treatment planning system for each patient and generating a 3D patient surface defined at a specified Hounsfield unit threshold of -200 HU. The topographic surface dose maps were then sampled downward from the surface with an exponential depth weighting to 1 cm, and then applied as textures to the 3D surface. For the Cherenkov imaging, camera calibration was performed with a standardized checkerboard pattern, allowing for visualization of this surface and dose maps from the camera field of view for quantitative comparison with imaged Cherenkov intensity distributions. For each fraction imaged from two cameras, Dice coefficient (% similarity), MDC (mean distance to conformity), and incident coefficient (confidence of signal outside the planned areas) were measured by comparing predicted surface dose to the Cherenkov images. Additionally, all images were examined visually to describe differences seen between these two images. <h3>Results</h3> Between May-June 2021, Cherenkov images were taken of 204 fractions and surface planning volumes from 21 plans from 21 patients. Comparing the treatment plans to the Cherenkov images, MDCs were found to average 3.4 mm (std dev 1.8mm). The mean DICE coefficients averaged 93.7% (std dev 2.8%). The mean incident coefficient averaged 0.4% (std dev 0.6%,). Visually, the plan overlay was identical to the imaged Cherenkov dose in 14 of the patients. For the remainder of the patients, there were minimal areas of non-overlap noted on four patients, and moderate areas of non-overlap in three patients. Areas of non-overlap were easily identified including lateral, medial, and superior edges of planed fields. <h3>Conclusion</h3> Advancements in Cherenkov imaging allows us to compare fractions with planned surface dose. The data is encouraging for using the predicted plan as the gold standard of comparison to the daily Cherenkov images. By comparing all fractions to an initial planned surface dose, we avoid potentially using a non-ideal fraction as a comparator and thereby introducing systematic bias. Future work will focus on integrating this new gold standard into advancements in an automated notification system for detecting incidences by Cherenkov imaging.