Impact of Respiratory Motion on the Image Quality of Fast and Slow Thoracic Hypersight CBCT: Implications for Online Adaptive Radiotherapy
Abstract
Purpose
To quantify the effects of respiratory motion and target size on internal target volume (ITV) geometric accuracy and Hounsfield Units (HU) stability in fast (6 s) and slow (60 s) thoracic HyperSight CBCT acquisitions.
Methods
A QUASAR respiratory phantom with a cedar insert including four spherical soft-tissue targets (0.5, 1, 2 and 3.0 cm diameter) was scanned under different sinusoidal motion (5, 10 and 15 mm amplitudes; 3, 4 ,6 and 7.5 s periods). Reference imaging included 4D fan-beam CT generating maximum intensity projection (MIP) images for ITV definition and time-averaged CT for HU benchmarking, all contoured in MIM software. HyperSight CBCT was acquired in 6 s and 60 s thoracic modes using the iCBCT Acuros reconstruction algorithm. Dice similarity coefficients and mean HU differences relative to time-averaged 4DCT were analyzed across motion conditions using paired t-tests (p-value).
Results
Under static conditions, HyperSight CBCT demonstrated minimal intrinsic Variation, with target HU differences within ±10–15 HU and Dice coefficients ≥ 0.8 relative to fast scan from CT-simulator. Under respiratory motion, geometric and density errors increased with decreasing target size and increasing motion amplitude. Dice coefficients were significantly (p-values ≤0.02) higher for slow compared to fast acquisitions for the 3 cm (0.89 vs 0.86), 2 cm (0.83 vs 0.80), and 0.5 cm targets (0.51 vs 0.42), but not for the 1 cm target. Cedar regions showed greater HU underestimation in slow scans versus fast scans (−35.5 vs −27.2 HU). Target HU discrepancy was significantly larger for slow scans in the 2 cm targets (p-value=0.03), while the smallest targets exhibited large variability and reduced fidelity.
Conclusion
Respiratory motion limits the geometric and quantitative reliability of thoracic HyperSight CBCT, with slow scans increasing HU error and fast scans introducing geometric variability. For targets ≤1 cm, both modes degrade accuracy.