1 INTRODUCTION

Since the beginning of the 21st century, cone-beam computed tomography (CBCT) has progressively established as important part of the imaging workflows in several medical fields,1-9 such as dentistry,1 orthopedics,2 image-guided radiotherapy,3, 4 and brachytherapy.5 Due to their often mobile configuration,6-8 corresponding CBCT-scanners are particularly suited and reveal high potential for both intraoperative6, 7 and interventional8 applications.

Modern CBCT-devices are equipped with a digital flat-panel detector (FPD) and cone-beam emitting X-ray source, both mounted to the system's gantry. With this setup, planar images of the entire anatomical region of interest (ROI) are acquired from several hundred to thousand different projection angles, forming the basis for the reconstruction of a 3D field of view (FOV).9 The resulting CBCT-images are in general characterized by high isotropic spatial resolution in the submillimeter range9, 10 with appropriate geometric accuracy.10, 11 Drawbacks, especially compared to conventional multi-slice computed tomography (CT), are mainly increased image noise, reduced image uniformity, as well as inferior low-contrast differentiability.9, 12 Respective image quality improvements are subject of ongoing research.13-15 For instance, Sheth et al.13 demonstrated improved low-dose performance with respect to noise, resolution, and detective quantum efficiency using complementary metal-oxide-semiconductor (CMOS) detectors instead of common SI:H-based FPDs. Jin et al.14 substantially reduced the amount of image artifacts and noise by combining a direct reduction and respective software-sided correction of scatter radiation. Kida et al.15 increased uniformity and signal-noise-ratio using a deep convolutional neural network.

For image-guided adaptive brachytherapy, the novel, mobile CBCT-system ImagingRing m (IRm; medPhoton, Salzburg, Austria) was recently deployed at our hospital as worldwide first site. The IRm features a 43.2 × 43.2 cm2 FPD, 121 cm gantry clearance, battery-powered maneuverability, and full remote control via a tablet-personal computer (PC). This ensures high flexibility and mobility, facilitates both interventional and intraoperative imaging, and distinguishes the device from previous X-ray systems.

Prior to its clinical operation, a profound characterization of the IRm's imaging performance was essential. This procedure allows, by assessing objective image quality and dose parameters, to draw conclusions about required adaptions of existing imaging workflows and/or the device's performance to underlying clinical requirements. Comparable characterizations of previous X-ray systems6, 7, 13, 16 included, for instance, the assessment of noise characteristics, contrast behavior, spatial resolution, and uniformity as fundamental image quality parameters.

The scope of the present work is the technical assessment of the CBCT imaging performance of the IRm. For this purpose, both scan protocols provided by the manufacturer as well as in-house protocols adapted to clinical requirements were used. In addition to a standard quality assurance (QA) phantom for determining physical imaging parameters, anthropomorphic phantoms were examined for evaluating the subjective quality of scans of different anatomical ROIs.

2 MATERIALS AND METHODS 2.1 ImagingRing m

The IRm is a novel X-ray system featuring CBCT, radiography, and fluoroscopy as imaging modalities. It is built upon an aluminum ring gantry with 121 cm clearance, on which an X-ray source and FPD are mounted independently movable (Figure 1). Both source and detector can rotate about more than 650° around the gantry's rotational axis driven by cable pulls (thus being able to perform, depending on start/end-positions of source and detector, 2–3 scans without having to change the rotation direction) and be arranged at any angular position with an accuracy of 0.1°. The individual positioning of source and detector enables non-isocentric as well as stitched volumetric and planar imaging. Technical specifications of gantry, source, and detector are given in Table 1.

Shown is the technical setup of the IRm (a), which is constructed from an X-ray source and flat-panel detector (FPD) mounted on a gantry with 121 cm clearance and controlled via the portable Human Machine Interface (HMI). Motorized wheels integrated to the system's legs ensure high mobility. Particularly, an Alderson® upper torso as well as a CIRS breast phantom (b), in which plastic catheters were implanted, were used for imaging assessment of the IRm. (c) A zoom image of the breast phantom is shown

TABLE 1. Technical characteristics of the IRm's gantry, source, and detector Geometrical construction Physical dimensions 182 × 87 × 190 cm3 Gantry bore 121 cm clearance Source axis distance 74.3 cm Source detector distance 126 cm Orbital gantry range About 650° Source characteristics Generator HF1 GMX-350/S2 (IMD) Anode RTM 780H 0.3/0.6 (IAE) Focal spot 0.3 mm (max. power: 6 kW) 0.6 mm (max. power: 25 kW) Tube current 0.01–120  mA Tube voltage 60–120 kV (in steps of 10 kV) Pulse length 2–35 ms Pulse rate Up to 30 Hz Prefiltering Air; 1.5 mm Al; pelvis Cu-Bowtie; 0.2, 0.5, and 1.5 mm Cu IRm's inherent filtration 4.4 mm Al equivalent (at 75 kV) Detector characteristics Model XRD4343RF (Varex Imaging) Active area 43.2 × 43.2 cm2 2880 × 2880 pixels Pixel pitch 150 μm (1 × 1 binning) Binning modes 1 × 1, 2 × 2, 3 × 3 binning Read-out frame rate 12 fps Scintillator Direct deposit CsI:TI Detector array Single substrate amorphous silicon active TFT diode array

For generating an X-ray cone-beam, the IRm is equipped with a HF1 GMX-350/S2 (IMD, Grassobio BG, Italy) generator with integrated anode of type RTM 780H 0.3/0.6 (IAE, Cormano Milano, Italy). This setup enables imaging with two focal spots of 0.3 mm (maximum tube-current: 30 mA, maximum power: 6 kW) and 0.6 mm (120 mA, 25 kW) size. Examinations can be performed at tube voltages of 60–120 kV (in 10 kV steps) with both continuous or pulsed (pulse lengths: 2–35 ms) tube output.

The IRm provides a fixed inherent minimum filtration equivalent to 4.4 mm aluminum (at 75 kV). For additional spectral hardening, an independently adjustable filter wheel and carriage are integrated within the beam path of the device. The filter wheel contains air, 1.5 mm Al, 0.2 mm Cu, and 0.5 mm Cu as prefilters. The filter carriage comprises air, 1.5 mm Cu, as well as a Bowtie filter specified for the pelvic region. Moreover, within a so-called volume-definition workflow (VDW), four independently movable jaws serve to dynamically collimate the X-rays towards the actual anatomical ROI only. Within this VDW, an anterior–posterior (AP) and lateral topogram is acquired, based on which the 3D FOV can be specified prior to a CBCT acquisition. The jaws enable for each projection a maximum planar isocentric FOV of 25.4 × 25.4 cm2.

For image acquisition, the IRm features a XRD4343RF (Varex Imaging, Salt Lake City, UT, USA) FPD, which consists of a direct deposit CsI:Tl scintillator that is connected via fiber optic plate to a single substrate amorphous silicon active thin-field transistor diode array. The detector has an active area of 43.2 × 43.2 cm2 and can be operated in 1 × 1 (pixel pitch: 150 μm, 2880 × 2880 pixels), 2 × 2, and 3 × 3 binning modes. It is currently read out with 12 Hz frame rate.

The operation of the IRm is performed via WiFi-based remote control by means of a portable so-called Human Machine Interface (HMI, Figure 1), which is established on a Windows tablet-PC equipped with additional joysticks and buttons. As visualized in the supplementary materials, motorized wheels integrated into the system's legs (Figure 1) as well as gearboxes enable longitudinal and lateral translation movements, free rotations on the floor, and up to ±30° gantry tilt. All maneuvers can particularly be carried out in a battery mode for up to 30 min. Manually moving the device without the HMI is currently not possible in standard operation. However, for emergency situations such as battery failure, emergency wheels can be extended from the legs via hand cranks, which allow the IRm to be moved manually in longitudinal direction.

Four system cameras, two integrated in the gantry and two in the detector, allow patient monitoring during ongoing examinations. The cameras are particularly suited also for real-time tracking of medical instruments, for example, for image-guided surgery. In addition, built-in lasers serve to align the system for scans and to visualize the FOV of examinations directly on the patient's skin. This facilitates both the exact positioning of the device and the imaging workflow.

2.2 Scan protocols

The scope of the present work is the assessment of the CBCT imaging performance of the IRm. Both standard scan protocols provided by the manufacturer and protocols developed in-house were used. In particular, in-house protocols for breast and pelvis imaging were defined based on our CBCT experience, as these anatomical regions represent our main clinical focus for image-guided adaptive brachytherapy and the corresponding manufacturer's protocols did not fulfill our clinical requirements. The scan parameters of the protocols investigated in this work are listed in Table 2. In the following, for brevity, all protocols are referred to by the abbreviations listed in Table 2. Note that the IRm does not feature any automatic exposure control, which is why image acquisitions generally have to be explicitly adapted to different patient sizes/characteristics: for the in-house protocols, all parameters were determined based on experience; for the manufacturer protocols, the parameters resulted automatically from selecting different dose levels (low, medium, high) within the IRm's control software.

TABLE 2. Listed are the main scan parameters (scan time, tube voltage, current-time-product per frame prior to prefiltering, number of frames, prefiltering, reconstruction kernel with frequency cutoff as fraction of the Nyquist frequency in brackets, velocity modulation (VM), and focal spot (FS) size) of all examined protocols Parameters Protocol Time (s) Voltage (kV) Tube output (mAs) Frames Prefilter Kernel VM FS (mm) Custom protocols Pelvis - Low dose (PL) 20 120 0.60 240 0.2 mm Cu CO (0.9) – 0.3 - Standard (PS) 25 120 1.16 300 0.5 mm Cu SL (0.8) 2.5:1 0.6 - Obese (PO) 25 120 1.50 300 0.5 mm Cu SL (0.8) 2.5:1 0.6 - High-quality 360° (PH) 50 120 1.80 600 1.5 mm Cu SL (0.8) 2.5:1 0.6 Breast - Standard (BS) 17 110 0.90 204 0.5 mm Cu SL (0.8) – 0.3 - Obese (BO) 17 120 0.96 204 0.5 mm Cu CO (0.9) – 0.6 - High-quality (BH) 30 120 1.50 360 1.5 mm Cu SL (0.8) – 0.6 Standard presets Abdomen - Low dose (AL) 26 100 0.09 312 Air RL 2.5:1 0.3 - Standard (AS) 26 100 0.56 312 Air RL 2.5:1 0.6 - Obese (AO) 26 100 2.10 312 Air RL 2.5:1 0.6 Note: Both in-house protocols for pelvis and breast as well as manufacturer protocols for abdomen imaging were investigated. A Cosine (CO), Shepp–Logan (SL), and RamLak (RL) kernel were used.

The IRm can perform both short scan (orbital range: 180° + beam divergence) and full scan (360°) trajectories. In this work, we acquired all scans using the VDW and short scans with source right orbit. These allow lower acquisition times (at least 12.5 s) than full scans (at least 22.5 s) and, at fixed scan time, improved angular sampling. Only for the pelvis protocol PH, with the intention of image quality enhancements by avoiding potential short scan artifacts, a full scan trajectory with extended acquisition time was resorted to. For both breast and pelvis, the so-called high-quality protocols BH and PH referred to the use of the strongest prefiltering (1.5 mm Cu). Due to the high absorption of this prefilter, BH and PH featured a comparatively increased current-time-product and a significantly extended acquisition time, which also improved angular sampling. With the exception of PL, all pelvis protocols exhibited 2.5:1 velocity modulation (thus reducing the travel speed of source/detector in the lateral angular range by a factor of 2.5 compared to the AP range), to increase the overall dose rate in the lateral regime.

The manufacturer's protocols for the abdomen as third anatomical ROI were adopted directly from the IRm's control software without any adaptions. These were characterized in particular by using air prefiltering and 2.5:1 velocity modulation.

Image reconstruction is based on a variation of the Feldkamp–David–Kress algorithm,17, 18 starts already during running examinations in a so-called real-time reconstruction, and typically finishes within less than 50 s after scan completion. For the in-house protocols, a Shepp–Logan reconstruction kernel with cutoff 0.8 (fraction of the Nyquist frequency) was used in general. Only for the entity-specific protocols yielding the highest noise magnitudes in patient examinations (PL and BO), a Cosine kernel with cutoff 0.9 was utilized to reduce noise by slightly enhanced image smoothing. The manufacturer's abdomen protocols featured a RamLak kernel with cutoff 1.0. It has to be mentioned that the IRm's control software currently does not allow repeated (re-)reconstructions of acquired scans, and thus a retrospective adjustment of image parameters such as kernel or slice thickness is not directly possible. In general, the IRm is able to perform reconstructions using a RamLak, Shepp–Logan, Cosine, Hamming, or Hann kernel each with manually adjustable cutoff frequency.

All scans were acquired using the 2 × 2 detector binning. Within the reconstruction process, the scans were reconstructed initially with a voxel size of 0.2 × 0.2 × 0.5 mm3 and the re-binned to the actual set voxel size of 0.4 × 0.4 × 1 mm3 output to the user. The FOV had dimensions of about 20 × 20 × 20 cm3.

2.3 Geometric reproducibility

The geometric accuracy of CBCT-scanners is particularly subject to the influence of mechanical/gravitational sag and flex of scanner components such as source and detector. These intrinsic properties might cause deviations between the center of actual imaging and reconstruction, resulting in degradations of image contrast as well as misalignment artifacts.19-21 Consequently, there is a need to account for geometric inaccuracies in the scan trajectory and thus to calibrate the scanner geometry accordingly.

For geometric calibration of the IRm, the flexmap approach21, 22 is pursued by using a cylindric nine-degrees-of-freedom ball bearing calibration phantom21 that is supplied together with the IRm. The phantom is placed on the patient table after consultation with the manufacturer, fixed to a stable holder to prevent rolling. By means of this phantom, three (one for each spatial dimension x, y, z) translation corrections tCx, tCy, tCz and tSx, tSy, tSz of the detector midpoint C and focal spot S, respectively, as well as three Euler rotation angular corrections rx, ry, rz of the detector's row-/column-vectors are determined as function of the gantry angle. This is done as described in detail by Keuschnigg et al.21 and within the IRm's default calibration method following the manufacturer's standard operating procedure. As defined by this group, the x-component refers to the gantry's rotation direction, y to the direction along the rotational axis, and z to the radial direction within the tomographic plane. The calibration procedure involves a gantry angular range of 520° with 53 projection views in 10° steps. Clockwise and counter-clockwise gantry rotations are considered separately. The objective of the calibration is to ensure geometric accuracy in the reconstruction by applying the determined corrections angular-dependent to each projection of a scan.

The geometric reproducibility of the IRm was validated by performing the described procedure eight times in direct succession and analyzing the angular-dependent variations of the obtained corrections. The consistency of the individual measurements was evaluated by calculating both the mean offsets and the standard deviations of the results. The impact of the gantry's rotation direction was assessed by calculating the differences of the corrections obtained for clockwise and counter-clockwise gantry rotation for each calibration.

2.4 Physical imaging parameters

For assessing the CBCT image quality, the CatPhan® 504 (CatPhan; The Phantom Laboratory, Salem, NY, USA) was used. This is a modular phantom consisting of four individual segments, a detailed description of which is provided by the manufacturer.23 The phantom was placed isocentrically within the scanner on a carbon fiber table, unless otherwise mentioned, and served for evaluating imaging fidelity, CT-number accuracy, contrast-noise-ratio (CNR), noise characteristics, uniformity, and spatial resolution. Within the frame of this work, all considered image metrics were measured eight times each.

2.4.1 Imaging fidelity

In addition to the geometric reproducibility (Section 2.3), the imaging fidelity is of high importance for evaluating a scanner's geometric accuracy. In other words, it must be ensured that the real lengths of examined objects are accurately reflected on the scans. To quantify the IRm's imaging fidelity, we considered the four rods of the CatPhan module CTP404, which are arranged as square with side length 50 mm (Figure 2a). On the module's central axial slice, the Euclidean distances between the centers of the individual rods, that were determined via threshold-based detections, were measured pairwise. For each scan protocol, the absolute difference between the measured lengths and the real rod distances was averaged over all rod pairings (i.e., both the side and diagonal lengths of the square were considered) and all scans. This provided a quantitative measure of the geometric accuracy of the device. The standard deviation of the obtained differences corresponded to the associated geometric imaging uncertainty.

The CatPhan modules CTP404 (a), CTP486 (b), and CTP528 (c) used for physical image analysis. For uniformity assessment, computed tomography (CT)-number profiles were drawn within the CTP486 along the indexed directions (b). The four rods used for the assessment of imaging fidelity are labeled with the numbers 1–4. Images from our archive, acquired with the conventional CT-system SOMATOM go.Open Pro (Siemens Healthineers, Erlangen, Germany)

2.4.2 CT-number accuracy

CT-number accuracy was assessed by considering the seven inserts of the CatPhan module CTP404 (Figure 2a), embedded in a uniform background material.23 On the central axial slice of this module, a circular ROI was centered within each insert and the respective mean CT-number was measured.

2.4.3 Contrast-noise-ratio Based on the central axial slice of the CTP404 module, the CNR of the low density polyethylene (LDPE, similar CT-number to fat), polystyrene (soft tissue), and Delrin® (bone) insert in relation to the module background was calculated. These inserts were chosen, since they represent a broad human-body like CT-number spectrum. As reported in detail previously,6 the CNR was computed based on the CT-number mean ROIinsert and standard deviation σinsert of a circular ROI centered within each insert as well as ROIbkg and σbkg of an adjacent background ROI: (1) 2.4.4 Noise characteristics Furthermore, we characterized both magnitude and spectral composition of observed image noise. For this purpose, the 2D noise power spectrum (NPS) as function of the spatial frequencies fx,y (for both in-plane image directions x and y) was calculated for each scan. Out of two central slices of the homogeneous CatPhan module CTP486 (Figure 2b), a difference image was computed. The two slices were spaced four times the slice thickness to avoid cross-correlation effects. An ensemble of ROIs ( > 200) of size 5 × 5 cm2 was uniformly distributed within the CTP486 on this difference image, and the 2D NPS was obtained as proposed by Steiding et al.10 (for more in-depth description of the calculation, please refer to this original article): (2)

As described by this group,10 Δx, Δy correspond to the in-plane pixel sizes and Nx, Ny correspond to the numbers of pixels of a ROI in x- and y-directions, respectively. The discrete Fourier transform (DFT) of the ROIs , that were offset-corrected with their respective mean CT-number , allows the examination of the spectral composition of image noise in the frequency domain. The factor 1/2 accounts for the artificial noise increase induced by the difference image approach.10

For visualization, the 1D NPS NPS(fr) (with ) was computed by radial averaging of the 2D NPS. Discrete integration of the 1D NPS yielded the magnitude σNPS of the image noise.

2.4.5 Spatial resolution Spatial resolution of scans was assessed by considering a non-resolvable line pair structure, that was slightly rotated within the axial plane, of the CatPhan module CTP528 (Figure 2c). Several line profiles were drawn across this structure, and by subsequent appropriate superposition an oversampled edge-spread function was created.24 The latter was fitted with a Fermi edge function and derived to obtain the line spread function LSF(x), which was resampled to one-tenth of the pixel size of the scan. By DFT, the modulation transfer function MTF(fx) dependent on the spatial frequency fx was calculated: (3)

The MTF was normalized to MTF(0).

2.4.6 Uniformity

For evaluating image uniformity, 10 image rows and columns each were centered around the axial midpoint of the homogeneous CTP486. The CT-numbers of the corresponding pixels were averaged row- and column-wise, respectively, to create a mean CT-number profile for both the AP and lateral image direction. Emerging non-uniformities were classified as the maximum of the absolute CT-number differences between both profile edges (defined to be at 5% and 95% profile length) and the profile center. This methodology particularly accounts for potential asymmetric CT-number profiles. In addition to the isocentric placement of the CatPhan within the scanner, the phantom was also incrementally displaced along the gantry's rotational axis in 2 cm steps. This served to evaluate uniformity as function of the longitudinal distance from the reconstruction center.

2.4.7 Dose assessment

Dose measurements were performed using the IEC 60601-2-4425 compliant standard body dosimetry phantom. The phantom has a diameter of 32 cm, is made of acrylic glass, and was placed isocentrically within the scanner on a carbon fiber table. It features one central and four peripheral drill holes, into which a 10 cm pencil ionization chamber of type 30009 (PTW-Freiburg, Freiburg, Germany) was sequentially inserted for measuring the corresponding dose length products (DLPs). The manufacturer calibrated the detector with a reference radiation quality of 120 kV and 8.4 mm aluminum half-value layer and specified its energy response to ≤5% for tube voltages of 70–150 kV. Unused drill holes of the phantom were filled with acrylic rods.

Based on measured in the central drill hole and as average of the DLPs obtained in the four peripheral holes, the weighted cone-beam dose index (CBDIw)26, 27 indicating the mean point dose measured over the chamber length was calculated: (4)

The CBDIw was measured three times for each investigated scan protocol.

2.5 Anthropomorphic phantom studies

In addition to the evaluation of physical parameters, the imaging performance of the IRm was also assessed using an anthropomorphic upper torso Alderson® phantom (dimensions: AP = 24 cm, lateral = 31 cm, cranial-caudal = 72 cm; Radiology Support Devices, Long Beach, CA, USA). The phantom comprises, embedded into soft tissue simulating Rando-plastic, an artificial adult skeleton as well as integrated intestine, bladder, and lung structures and imitates a human thorax, abdomen, and pelvis.

Furthermore, 15 flexible plastic catheters (Type 6F; Elekta, Veenendaal, Netherlands) were implanted into an anthropomorphic breast phantom (Breast Elastography Phantom Model 059; CIRS, Norfolk, VA, USA) by using brachytherapeutic guidance needles. The catheters were fixed to the breast's skin with appropriate plastic buttons to prevent slippage. The prepared breast was placed on the thorax of the Alderson® phantom to simulate the female anatomy during breast examinations (Figure 1).

The visual assessment of the CBCT-scans allowed to evaluate the IRm's imaging performance in simulated clinical patient examinations. The acquired scans were interpreted with respect to the measured physical imaging parameters.

3 RESULTS 3.1 Geometric reproducibility and accuracy

The three focal spot and six detector corrections obtained for the clockwise gantry rotation direction in all eight calibrations are shown in Figure 3a–c. For each single calibration, the individual corrections varied along the entire angular gantry orbit as a result of mechanical inaccuracies, vibrations/oscillations, and/or gravity effects. The mean value of all corrections averaged over the entire gantry orbit and all eight calibrations was, except of for tCy, subject to systematic offsets different from zero. This offset was up to 11.5 ± 0.4 mm (tCz) for the translational corrections and up to 0.64 ± 0.02° (rx) for the rotational corrections. Similar systematic offsets were found for the counter-clockwise rotation direction, and are therefore not shown for brevity.

The three translational corrections of focal spot (a) and detector midpoint (b) as well as the detector rotation corrections (c) obtained for clockwise gantry rotation in all eight calibration procedures. Furthermore, the differences of the corrections obtained for clockwise and counter-clockwise gantry rotation were calculated for each calibration and then averaged per angle over all eight measurements. The absolute values of these averages are illustrated in (d-f), where no error bars are shown for clarity

As can be seen from Figure 3a–c, the calculated corrections varied considerably between the eight calibrations, even at fixed angles. For quantification, the standard deviation of the results obtained in all eight calibrations was calculated for each correction at each angle. The maximum angle-specific standard deviations ranged from 1.0 mm (tSy) to 5.3 mm (tSz) for the translational corrections and from 0.06° (rz) to 0.61° (ry) for the rotational corrections. Averaged over the entire orbital range, the mean standard deviations ranged from 0.6 ± 0.2 mm (tSy) to 2.1 ± 0.7 mm (tCz) and 3.0 ± 0.9 mm (tSz) for the translational corrections and from 0.039 ± 0.011° (rz) to 0.23 ± 0.09° (ry) for the rotation corrections, respectively. Especially the results for tSz and ry appeared to be unreproducible as shown in Figure 3a,c. Again, similar results were obtained for counter-clockwise gantry rotations.

In particular, the described high fluctuations were not exclusively caused by uncertainties in the software-sided calibration routine, but by actual geometric instabilities of the IRm. This was evident by considering eight scans of the CatPhan module CTP486 performed in direct succession with the full scan protocol PH. On the inner margin of the module, significant double structures appeared. These, however, were not stable (which could be an indication of a wrong geometric calibration only), but varied significantly between the eight scans as shown in Figure 4. For short scans, these effects were much less pronounced. This observation provides evidence for fluctuating geometric properties of the scanner and thus for existing geometric instabilities.

Double contours observed at the inner CTP486 edge, exemplary presented for the region marked black on the left image. The appearance of the double contours varied as illustrated between the eight considered PH scans acquired in direct succession, thus indicating geometric instabilities varying from scan-to-scan. Windowing—Level: -40 HU, Width: 500 HU

Furthermore, for each calibration procedure and angle, the difference between the corrections obtained for clockwise and counter-clockwise rotation direction was calculated and then averaged over all eight measurements. The absolute values of these mean deviations are shown in Figure 3d–f. Specifically for tSz, tCz as well as ry, highly fluctuating deviations between the two rotation directions were found. The deviations were up to 4.6 mm (tSz), 3.1 mm (tCz), and 0.3° (ry).

The observations that instabilities occurred particularly with the full scan protocol PH were confirmed by the examinations of imaging fidelity. For each protocol, the respective measurement results are provided in Table 3. Only minor differences were found between all short scan protocols. For these, the strongest geometric inaccuracy was measured for PS and amounted to 0.21 ± 0.18 mm. However, for the full scan protocol PH a significantly increased inaccuracy and especially measurement uncertainty of 0.39 ± 0.53 mm was obtained. This magnitude was roughly consistent with the appearance of the double structures shown in Figure 4. PH thus showed a worse imaging fidelity compared to all sh