HyperSight improves CBCT dose calculation and registration, but automated contouring remains a barrier

Across 323 treatment CBCT scans, HyperSight improved image completeness and dose agreement compared with iterative CBCT, while contouring reliability remained inconsistent.

KEY POINTS

  • This retrospective multicenter-department technical study evaluated 40 patients treated across breast, thoracic, abdominal, and male pelvic sites using either Varian Halcyon HyperSight (HypS) or TrueBeam iterative CBCT (TB-iCBCT). Investigators analyzed 323 treatment CBCT datasets, comprising 208 HyperSight and 115 TrueBeam images, and compared them with planning CT for automated contouring, deformable registration, electron-density accuracy, and dose calculation.
  • HyperSight's larger field of view reduced major or complete anatomical truncation by approximately eight percentage points compared with conventional iterative CBCT. Improvements were especially pronounced in breast treatment geometry, where the frequency of major or complete truncation decreased by 59% for the contralateral breast and 50% for the contralateral lung. HyperSight also allowed a larger proportion of the calculated dose distribution to remain within the acquired image volume (99.6% versus 90.9%).
  • Automated contouring consistency improved for 13 of 21 evaluated anatomical structures, when comparing contours generated independently on CBCT and planning CT. HyperSight achieved a majority of acceptable contour comparisons for 15/21 structures, compared with 12/21 for TrueBeam iCBCT. However, several low-contrast structures, including the pancreas, stomach, and seminal vesicles, remained problematic. Importantly, agreement was measured between automated contours rather than against a physician-defined anatomical ground truth.
  • Deformable registration generally achieved higher geometric agreement with HyperSight, although results were organ-dependent. In breast and thoracic anatomy, average post-registration Dice similarity was approximately 0.929 for HyperSight versus 0.875 for TrueBeam iCBCT. In abdominal and pelvic regions, corresponding mean values were 0.817 versus 0.798. The liver and selected other structures did not consistently favor HyperSight, indicating that improved image quality does not guarantee better registration for every organ.
  • Electron-density representation was more stable with HyperSight in heterogeneous anatomy. The mean deviation from planning-CT-derived physical density was −0.010 ± 0.043 g/cm³ for HyperSight, compared with +0.005 ± 0.116 g/cm³ for TrueBeam iCBCT. Agreement was similar in the pelvis, while HyperSight generally demonstrated smaller variations in thoracic regions. Density comparisons were influenced by anatomical changes and exclusion of structures with insufficient contour quality.
  • Dose calculation agreement with planning CT was high for both systems, with mean 3%/2 mm gamma passing rates of 99.5% ± 1.0% for HyperSight versus 96.8% ± 2.9% for TrueBeam iCBCT. HyperSight improved average gamma agreement by approximately 11.1 percentage points under 1%/1 mm criteria, with particularly favorable results in breast and prostate cases. However, the comparative advantage was less consistent for lung anatomy, and TrueBeam performed better in some abdominal cases because of differences in anatomical heterogeneity.
  • Accurate automated contouring remained the main obstacle to fully unsupervised offline adaptation. The commercial auto-contouring algorithm had been trained on planning CT rather than CBCT, and manual contour correction was deliberately excluded from the automated workflow assessment. The study also compared different patient cohorts rather than paired acquisitions on the same patients, did not perform prospective adaptive replanning or treatment, and excluded some datasets because of substantial image truncation or poor contour quality. The findings therefore establish component-level feasibility rather than a clinically validated automated decision system.

CLINICAL TAKEAWAY

HyperSight CBCT provides improved image coverage, electron-density stability, registration performance, and dose calculation accuracy that may support more efficient offline adaptive radiotherapy workflows. However, reliable anatomical contouring and patient-specific failure detection remain essential, and the retrospective nonpaired comparison does not establish that adaptation can safely proceed without clinical review.

SOURCE

Journal of Applied Clinical Medical Physics

Browse more research Suggest a correction