As briefly noted in the Introduction, HyperSight represents Varian Medical Systems’s latest upgrade to CBCT imaging, integrating both hardware and software innovations to enhance image quality, acquisition speed, and dose efficiency.

From a hardware perspective, HyperSight incorporates an extra-large flat-panel detector with an active area of 86×43 cm², allowing a scan diameter of up to 54 cm (Table 1 and Fig. 1). This expanded field of view enhances anatomical coverage while maintaining high spatial resolution. The detector design includes reduced lag characteristics and an integrated anti-scatter grid, thereby minimizing detector-related scatter and enhancing image uniformity. Additionally, a high-capacity X-ray tube has been introduced, providing a two-fold increase in heat capacity over previous generations. This upgrade enables higher power output, increased X-ray intensity, and the ability to perform multiple scans in a shorter time without overheating, supporting more flexible and efficient imaging workflows.

On the software side, HyperSight employs a novel iterative reconstruction framework that replaces the Feldkamp-Davis-Kress (FDK) algorithm [12-15]. The process comprises two phases: first, an initial reconstruction using the Algebraic Reconstruction Technique with total variation regularization to suppress noise and preserve edges; second, a Penalized Likelihood reconstruction that statistically refines the image based on the Poisson nature of X-ray detection. The pipeline also incorporates several correction modules, including deterministic scatter correction, hardware scatter modeling via Monte Carlo simulation, MAR, and energy-dependent HU calibration. These enhancements improve HU consistency, reduce artifacts, and enable superior soft tissue visualization—features especially critical for adaptive radiotherapy.

Following the installation of the HyperSight imaging system on the Halcyon platform, a comprehensive IPA procedure was conducted to verify system functionality and imaging performance in accordance with the manufacturer’s specifications. All tests were conducted by a certified Varian Medical Systems service engineer using the standardized protocols outlined in STB 11-612 HyperSight and CBCTp Options.

The acceptance process began with verification of the X-ray generator output. The accuracy of the tube potential (kVp), current (mA), and exposure time (ms) was measured using a RaySafe X2 dosimetry system (RaySafe). Additionally, beam quality was assessed by measuring the half-value layer (HVL) measurements at 70 kVp and 100 kVp to ensure that the filtration and output characteristics met the required tolerances. Subsequently, system-level image performance was evaluated by assessing the kV detector panel’s characteristics, including imager sensitivity (gain), offset uniformity (dark field image), electronic noise (dark noise), and pixel defect rate. These tests provided baseline information on detector performance and image stability under standard acquisition conditions.

To evaluate CBCT imaging performance, QUART DVTvn phantom (Varian Medical Systems) scans were acquired under both head (100 kVp) and pelvis (125 kVp) modes. Image quality was assessed by examining HU calibration accuracy across multiple insert materials, geometric fidelity through distance measurements, uniformity across central and peripheral regions of the phantom, and high-contrast resolution derived from edge response profiles. All parameters were compared against the manufacturer’s acceptance criteria to confirm compliance with expected system performance.

To further evaluate the imaging performance of the HyperSight-enabled CBCT system beyond the IPA, additional phantom-based tests were conducted using a diverse set of commercially available phantoms. These evaluations aimed to compare the image quality between the conventional FDK-based reconstruction method against the newly introduced Acuros CTS reconstruction method, which incorporates deterministic scatter correction based on the linear Boltzmann transport equation.

Three types of phantoms were utilized in this analysis. Alongside the QUART DVTvn phantom, which was also used during the IPA and served as a reference standard for HU calibration and uniformity under controlled conditions, we employed the Catphan 604 phantom (The Phantom Laboratory), a widely used clinical reference for evaluating image quality parameters. We also included the MZ CUBE phantom (Paprika Lab), a commercially available, multipurpose QA phantom designed to consolidate various CBCT quality assurance tests—including HU constancy, uniformity, spatial resolution, and low-contrast detectability—into a single device. A photograph of all three phantoms is shown in Fig. 2.

All phantoms were scanned on the Halcyon platform using multiple acquisition protocols, including standard FDK-based and Acuros CTS reconstruction methods. For each phantom–protocol–reconstruction combination, three repeated CBCT acquisitions were performed. HU constancy, uniformity, and noise metrics were calculated as the average of the three measurements to ensure measurement stability and reduce scan-to-scan variability. The primary image quality metrics assessed were HU constancy across known materials, image uniformity across central and peripheral regions, and image noise, newly introduced in this study, quantified as the standard deviation of HU values within homogeneous regions of interest (ROIs).

For HU constancy assessment, reference HU values for each material were defined according to the manufacturer-provided material specifications for the Catphan 604 and MZ CUBE phantoms. For the Catphan phantom, the reference HU was selected as the nominal mid-range value reported in the manual (e.g., Air –1,000 HU, PMP (polymethylpentene) –200 HU, LDPE (low-density polyethylene) –100 HU, Polystyrene –35 HU, Acrylic 120 HU), based on the HU ranges provided by the manufacturer for each insert. For the MZ CUBE phantom, the reference HU for each material was defined as the midpoint between the minimum and maximum HU values specified in the manual (e.g., PE –47 HU, Acrylic 114.5 HU, Air –997 HU, Teflon 1,000.5 HU). These reference values served as the baseline for calculating HU constancy deviations across all reconstruction conditions.

For assessing image uniformity and noise, five ROIs were analyzed: one at the center of the phantom and four peripheral ROIs positioned 50 mm from the center along the superior, inferior, left, and right directions. In the Catphan phantom, each ROI was defined as a 20×20 mm square, whereas in the MZ CUBE phantom, circular ROIs of approximately 3 cm² were used, placed at the same radial distance of 50 mm from the center. These ROI geometries and placements adhered to the standard Catphan and MZ CUBE evaluation procedures. Table 2 summarizes the imaging parameters, reconstruction settings, and evaluation procedures used for each phantom and reconstruction method.

All tests conducted during the IPA procedure were completed and met the acceptance criteria specified in the manufacturer’s documentation. As outlined in the Methods section, the X-ray generator output was verified by evaluating the accuracy of kVp, mA, and ms, as well as by measuring the HVL at multiple energy levels. All measured values were within the specified tolerance ranges. In addition, the kV imaging panel was comprehensively evaluated for baseline image quality characteristics. Specifically, tests were conducted for imager sensitivity (gain), dark field images (offset), noise images (dark noise), and kV pixel defects using the system’s built-in calibration tools. The results confirmed that all parameters were within the manufacturer’s specified limits, indicating that the HyperSight system was functioning properly at the time of installation. Table 3 summarizes the detailed outcomes of each IPA test and their compliance with the acceptance criteria.

The first aspect of the CBCT image quality evaluation focused on HU constancy across different materials using the Catphan 604 and MZ CUBE phantoms. For both phantoms, all tested materials exhibited HU values within the predefined ±50 HU tolerance (Halcyon 4.0 Specifications), confirming acceptable quantitative accuracy under both reconstruction methods. Quantitatively, the mean absolute HU deviation for the Catphan phantom decreased from approximately 14.6 HU to 8.5 HU in head mode and from 8.9 HU to 4.0 HU in pelvis mode when switching from FDK to Acuros, demonstrating a clear and consistent improvement across most materials. For example, for the air insert, the deviation decreased from 19.9 HU to 5.7 HU under the head protocol and from 10.6 HU to 0.2 HU under the pelvis protocol. The air insert exhibited the largest difference between FDK and Acuros reconstructions because air generates the strongest scatter gradients and cupping artifacts in FDK reconstruction. As FDK does not model scatter, HU underestimation is most severe in air, whereas Acuros effectively corrects these errors through deterministic scatter correction and energy-dependent modeling. LDPE was the only material for which the difference between FDK and Acuros remained small. Because LDPE has the lowest density and effective atomic number among the polymer inserts, it generates minimal scatter and exhibits weak energy-dependent attenuation. As a result, FDK already reconstructs LDPE with little intrinsic bias, leaving limited scope for further HU improvement.

In the MZ CUBE phantom, a similar pattern of material-dependent variation was observed. HU deviations decreased modestly for PE and acrylic and more substantially for Teflon (e.g., 949.6 HU to 973.1 HU in head mode), while air showed a small but consistent improvement. The high-density Teflon insert is subject to stronger beam hardening and nonlinear attenuation at 100 kVp, resulting in a greater improvement with Acuros reconstruction in the head protocol than in the pelvis protocol. In addition, the air region in the MZ CUBE phantom is smaller and embedded within a more heterogeneous surrounding structure, which inherently results in reduced scatter-induced HU bias. Consequently, the relative improvement achieved with Acuros appears smaller than that in the Catphan phantom. These findings indicate that Acuros-based reconstruction improves HU fidelity for most materials and may be particularly advantageous in scenarios requiring high quantitative accuracy. Detailed HU comparison results for each material and reconstruction method are illustrated in Fig. 3.

The second evaluation focused on HU uniformity across multiple ROIs within each phantom scan. Across all measurement directions and phantom types, the Acuros CTS reconstruction consistently demonstrated superior uniformity compared with the conventional FDK method. Quantitatively, for the Catphan phantom, the mean peripheral–central HU deviation decreased from 5.3 HU to 2.9 HU under the head protocol and from 3.6 HU to 1.9 HU under the pelvis protocol, representing average reductions of approximately 2.4 HU and 1.7 HU, respectively. A more pronounced improvement was observed for the MZ CUBE phantom, where the mean deviation decreased from 12.9 HU to 5.3 HU in the head mode and from 32.6 HU to 3.1 HU in the pelvis mode, corresponding to reductions of 7.6 HU and 29.5 HU, respectively. In particular, the pelvis protocol showed a dramatic improvement across all peripheral ROIs (e.g., top ROI: 35.66 HU to 2.67 HU). These findings demonstrate that Acuros more effectively suppresses image nonuniformities than FDK, with notably pronounced benefits under the pelvis acquisition protocol. A complete comparison of HU uniformity results across reconstruction methods and protocols is presented in Fig. 4.

The final analysis assessed image noise by measuring the standard deviation of HU values within homogeneous regions of the Catphan 604 and MZ CUBE phantoms. In both phantoms and across all ROI positions, Acuros CTS consistently demonstrated lower image noise compared to FDK. In the Catphan phantom, the mean noise level across the five ROIs decreased from 13.7 HU to 10.0 HU, representing an average reduction of approximately 27%. For example, in head mode, the noise in the central ROI decreased from 14.6 HU to 10.4 HU. A similar pattern was observed in the MZ CUBE phantom, where the average noise level decreased from 7.8 HU to 5.4 HU, representing an approximate 31% reduction. A detailed comparison of noise measurements across all ROIs is presented in Fig. 5.

The superior HU consistency and uniformity observed with Acuros CTS can be attributed to fundamental differences between the two reconstruction approaches. The conventional FDK method relies on purely geometric back-projection and does not explicitly model X-ray scatter, beam hardening, or the statistical nature of photon detection. As a result, residual scatter and nonlinear attenuation effects introduce spatially varying HU bias, particularly in large or dense phantoms and in peripheral regions. In contrast, Acuros CTS incorporates deterministic scatter correction based on the linear Boltzmann transport equation, energy-dependent material modeling, and statistical iterative refinement. This enables more accurate correction of scatter-induced shading, reduces cupping artifacts, and stabilizes voxel-wise HU values across the field of view. Together, these features directly contribute to the improved HU uniformity and quantitative accuracy demonstrated in this study.