Session Item

Two algorithms for post-processing of uncalibrated CBCT images were evaluated for all patient groups treated with curative intent at Vejle Hospital, with emphasis on dosimetric accuracy and applicability for offline treatment adaptation.

CT and CBCT images of 60 patients were identified. Patients were distributed with 10 breast, 10 lung, 10 prostate, and 10 anal or rectal cancer patients with original treatment plans optimized on CT images, and additionally 10 prostate and 10 anal or rectal cancer patients with treatment plans optimized on synthetic MR-derived (MRCAT) images. Treatment planning and CBCT based recalculation was performed in RayStation (research build v10.1.110.51).
“Clinical CBCT” (clinCBCT) dose was calculated on CBCT images acquired on the Elekta XVI system for IGRT, after assignment of standardized bulk densities. Additionally, two synthetic CBCTs were created per patient. A “corrected CBCT” (corrCBCT), calculated from deformable image registration (DIR) and 2D joint histogram analysis, and a “virtual CT” (vCT) created by DIR of the original planning CT to the clinCBCT image while correcting for local variations such as bowel gas etc.
Dosimetric accuracy was evaluated through gamma pass rate analysis with 2%/2mm and 1%/1mm criteria, as well as by evaluating D2% to organs at risk (OAR) and D50% to delineated target structures (propagated to the CBCT images using DIR). The reference image for comparison was chosen as a reference CT (refCT) as close in time as possible (same day CT for 21 patients who were rescanned during the treatment course, and planning CT vs. first fraction CBCT for the remaining 39 patients), with additional DIR performed to minimize anatomical variations. 

Gamma pass rates and DVH metrics are shown in Table 1. For the clinCBCT images, Gamma 2%/2mm pass rates as low as 37% are observed, while the corrCBCT and vCT present pass rates no less than 93.6% and 98.3%, respectively. At the 1%/1mm criterion (Figure 1), the clinCBCT median pass rate for the lung patients is as low as 30.6% (range 18.1-47.9%), while corrCBCT images pass at 98% of points in median (range 91.4-99.4%), and the vCT images pass at 97% of points (median, range 91.3-99.1%).

DVH metrics for the clinCBCT images showed a very wide range in difference for target D50%, from -26.1Gy for a breast patient to +10.9Gy for a CT-based prostate patient. OAR D2% ranged from -4.4Gy for an anal/rectal patient, up to +13.5Gy for a CT-based prostate patient. For the corrCBCT and vCT images, all variations were less than ±1Gy (targets and OAR). None of the CBCT based dose distributions were consistently unskewed in Spearmans signed rank test.

corrCBCT and vCT images provide accurate dose calculations, highly suitable to assist any strategy for offline dose-based adaptive radiotherapy, without needs for individual CBCT calibration or limitations to certain anatomical sites, CBCT acquisition protocols, or planning CT modality.