Cadaveric specimens
We examined the elbow joints of eight, randomly chosen, formalin-fixed cadaveric specimens with the novel prototype cone-beam CT scan mode of a multi-use, twin robotic x-ray system (Multitom Rax, Siemens Healthineers, Erlangen, Germany) and a high-resolution multidetector CT scanner (Somatom Force, Siemens Healthineers, Erlangen, Germany). MDCT scans were conducted in a modified “superman” stance, while CBCT studies were performed using the tableside scan position for 3D imaging of the upper extremity. The scan postures for MDCT and CBCT elbow imaging are presented by a staff member in Fig. 1.
Technical parameters and dose assessment
The x-ray system tube and flat panel detector are mounted on two motor-driven, telescopic arms that are connected to ceiling rails. While independent arm movement facilitates two-dimensional and fluoroscopy imaging, both arms move concurrently along predetermined paths for the acquisition of 3D projection data in CBCT scan mode (Fig. 2). The input field of the flat panel detector measures 22.8 cm × 21.3 cm with a 3D image matrix of 1,540 × 1,440 pixels (pixel size 148 μm). The x-ray tube is capable of voltages from 40 to 150 kV and currents from 0.5 to 800 mAs.
The tableside trajectory used for elbow and wrist imaging has a sweep angle of 200° and an asymmetric source-to-image-distance of 115 cm. Due to an increase in maximum frames per second from 16 to 30, the total scan time is reduced with the new prototype from 20 to 12 s when compared to the commercially available software version. Taking into account the two telescopic arms acceleration and deceleration phase, 304 projection images are acquired during each scan. Dose levels and subsequently tube current can be varied for different scan protocols with an automatic dose modulation system keeping the detector dose level constant. Therefore, a sensor at the detector continuously measures the incoming radiation and the automatic exposure control modifies the tube current-time product accordingly to the preset dose values.
For this study, however, we decided to use constant tube currents to provide comparable circumstances for both imaging modalities. To compare the applied radiation dose between CBCT and MDCT, the dose-area product of the CBCT scan mode was multiplied by a linear scaling factor leading to volume computed tomography dose index (CTDIvol)-equivalent values. By using a polymethyl methacrylate dosimetry phantom, conforming to International Electrotechnical Commission 60601-2-44:2009 with a total length of 300 mm and a diameter of 160 mm as well as a standard dosimetry system (Nomex Dosimeter, PTW, Freiburg, Germany) with a 300 mm ionisation chamber, the scaling factor for each combination of voltage, pre-filtration and acquisition geometry was calculated in advance. To achieve this, the dose-length product (DLP) was quantified in five chambers. Then, the standard weighting scheme for dose measurements was applied to each value to determine the DLPvol values. CTDIvol values were calculated by dividing the DLPvol by the beam width (equivalent to the field of view in the z-direction), and to finally compute the scaling factor, CTDIvol was divided by dose-area product values. We operated with a dedicated low-dose (LD), i.e., CTDIvol(16 cm) = 3.3 mGy, and a regular dose protocol (RD) comparable to our clinical scan protocol, i.e., CTDIvol(16 cm) = 13.8 mGy), for both systems. MDCT images were acquired using the CT scanner in single-energy mode. Scan parameters for CBCT and MDCT studies are displayed in Table 1.
Image reconstruction parameters
For the multidetector CT system, scanner-side reconstruction of raw data was conducted using a dedicated bone kernel (Ur77; Siemens Healthineers, Erlangen, Germany). The CBCT data was reconstructed with a comparable prototype kernel that provides equivalent standardised resolution numbers in the axial plane according to vendor information. Multiplanar reconstructions (MPR) of the cadaveric elbow scans were then performed for CBCT and MDCT using special software (Syngo via, Siemens Healthineers, Erlangen, Germany). Irrespective of dose protocol or scanner, slice thickness of 1 mm, increment of 0.5 mm, image matrix of 1,024 × 1,024 pixels and field of view of 80 mm were chosen for axial, coronal and sagittal planes. We decided on preset values for window width and level of 3,000 and 1,000 Hounsfield units for optimal bone visualisation. However, readers were allowed to change window settings as needed.
Subjective image analysis
Two independent radiologists with nine (T.G.; Reader 1, R1) and seven (S.H.; Reader 2, R2) years of experience in musculoskeletal radiology evaluated all elbow studies using the Merlin picture archiving and communication system (Phönix-PACS, Freiburg, Germany). In the first step of their reads, observers reviewed all of the images for each MPR in randomised and blinded order. After blinded review of all data was complete, readers were asked to assess whether the presented image quality of each study was sufficient for diagnostic use and to rate the overall image quality on a seven-point Likert scale (7 = excellent; 6 = very good; 5 = good; 4 = satisfactory; 3 = fair; 2 = poor; 1 = very poor). Furthermore, artefacts and image noise in osseous and soft tissue were judged separately using a five-point Likert scale (5 = minimal artefacts or noise; 4 = little artefacts or noise; 3 = moderate artefacts or noise; 2 = considerable artefacts or noise; 1 = strong artefacts or noise).
Objective image analysis
Software-based estimation of the signal intensity distribution in osseous tissue was performed in order to objectify the task of image quality evaluation. First, the range of signal intensities was calculated for image parts containing bone tissue. Subsequently, the fraction of pixels with a signal intensity between 25 and 75% of the calculated range was computed. Assuming that pixels above the 75%-border equal trabeculae or cortical bone and pixels below the 25%-border represent fatty marrow, the pixels between the borders were deemed as “undecided”, representing a mixture of bone and marrow and partly arising from blurred edges between both. Consequently, a larger share of these pixels indicated impaired image quality. We used special numerical computing software (MATLAB Version 2019a, MathWorks, Natick, MA, USA) for objective evaluation of image quality.
Statistical analysis
Dedicated software (SPSS Statistics, Version 23.0 for Mac, IBM, Amonk, NY, USA) was used to carry out statistical analyses. We present categorical variables (e.g., scale results) as frequencies, percentages and medians, whereas normally distributed data is presented as means ± standard deviation. The normal distribution of continuous variables was assessed using Kolmogorov-Smirnov tests. Paired nonparametric variables were compared using Wilcoxon signed-rank tests. For quantification of interrater reliability, we report the intraclass correlation coefficient (ICC) based on the absolute agreement in a two-way random-effects model. The interpretation of ICC values was conducted according to Koo and Li [17] (< 0.50 = poor reliability; 0.50–0.75 = moderate reliability; 0.75–0.90 = good reliability; > 0.90 = excellent reliability). The p values ≤ 0.05 were considered to indicate statistical significance.