Author: Harald Keller, Iymad Mansour, Jeff D. Winter 👨🔬
Affiliation: Princess Margaret Cancer Centre 🌍
Purpose: The growing number of adaptive therapy applications is motivating image segmentation and direct dose calculation on CBCT. The purpose of this investigation is to evaluate the recently released TrueBeam HyperSight imaging system metallic artifact reduction (MAR) algorithm in terms of image quality improvement and HU accuracy compared to other computed tomography (CT) platforms.
Methods: Metallic artifact severity was assessed across three CT platforms: Canon Aquilion CT (CT-simulator), Varian ETHOS with HyperSight, and Varian TrueBeam with HyperSight (TBHyper). A Sun Nuclear Multi-Energy CT phantom with custom metallic inserts (density: 8.0 g/cc; diameters: 3–13 mm) was imaged using a standard pelvic imaging protocol with platform-specific MAR implementations. The metallic inserts were placed centrally and off-centre within the phantom. Image artifacts were qualitatively evaluated, and Hounsfield Unit (HU) recovery in the surrounding plugs was calculated relative to a reference configuration with water-equivalent inserts.
Results: All MAR implementations qualitatively suppressed metallic artifacts. The CT-simulator and ETHOS platforms improved HU accuracy in surrounding inserts. With a 13 mm centrally placed insert, maximum absolute differences across plugs reduced from 60 to 8 HU (standard and MAR respectively) for the CT-simulator and from 120 to 35 HU for ETHOS. However, TBHyper’s MAR worsened HU recovery (mean absolute differences: 13 vs. 35 HU, maximum absolute differences: 80 vs. 200 HU). TBHyper’s MAR algorithm corrects the artefacts in the immediate vicinity of the metal, but enhances the overall artefacts, resulting in an increased range of HU deviations. On the other hand, off-center placement of the insert reduced HU discrepancies for all platforms.
Conclusion: All platforms’ MAR algorithms qualitatively improved imaging with metallic inserts, however HU accuracy improvements varied. Larger metal inserts and central placement exacerbated HU deviations, particularly for TBHyper. Unlike ETHOS and CT-simulator, TBHyper’s MAR algorithm did not consistently improve HU accuracy.