No institutional review board approval was necessary for this prospective phantom study.
Phantom design
A custom idealised phantom was developed at our institution by filling three polyvinyl chloride tubes (diameter 27 mm) with approximately 20 mL of one of the three fluids: red marrow analogue (RMA), yellow marrow analogue (YMA), or water. Each tube was also partially filled with 7.5 mL of bone granulate (Fig. 1a). This resulted in six compartments: three compartments of fluid without bone and three compartments of fluid with bone.
For RMA, heparinised human full blood was used, while for YMA, sunflower seed oil was used. Bone granulate was extracted from the cancellous bone from the neck of the femur in young cows. The bone was rinsed and dried before being crushed using a coarse grinder. All tubes were arranged in line and submerged in a 12 × 7 cm water chamber. All six compartments were placed equidistant from the isocentre of the scanner (Fig. 1b), and the entire phantom fixed to the scan table.
Dual-energy scan series
Scans were performed on a third-generation dual-source dual-energy system (Somatom Force, Siemens Healthineers, Forchheim, Germany). All scans were performed in a single session in August 2018. Doses for each scan were estimated during the setup of the scan protocol at the scanner and expressed as the volume computed tomography dose index (CTDIvol).
In the first experiment used for calibration, the phantom was scanned at a constant CTDIvol of 80 mGy at all five available tube-voltage combinations (70 kV/Sn150 kV, 80 kV/Sn150 kV, 90 kV/Sn150 kV, 100 kV/Sn150 kV, and 80 kV/140 kV; Sn indicating the use of a 0.6-mm tin filter). In addition, a single scan was performed in a high-resolution mode. In the second experiment, the phantom was again scanned with all five available tube-voltage combinations. For each tube-voltage combination, the phantom was scanned with six tube currents adjusted to result in equal doses. The CTDIvol for the six scans was between 1 and 80 mGy. In the third experiment, the tube-voltage combination was kept constant, but the rotation time was varied between 0.25 s and 1 s. To keep a constant CTDIvol, the tube current was reduced correspondingly. In the fourth experiment, the pitch was varied between 0.3 and 1.2 with constant CTDIvol and a constant tube-voltage combination. In the fifth experiment, the phantom was scanned once but images were reconstructed with eight different quantitative kernels with increasing sharpness. The automatic tube current modulation was disabled for all scans. Exact scan and postprocessing parameters are listed in Additional file 1: Table S1.
Image postprocessing
CT series were postprocessed on a commercially available three-dimensional workstation (syngo.via version VB20-HF5, Siemens Healthineers, Forchheim, Germany) using the bone marrow oedema algorithm. Results from the first experiment were used to calibrate the settings of the three-material decomposition, by a single investigator. For each tube-voltage combination, scanned at CTDIvol of 80 mGy, mean HU values of YMA and RMA without bone were measured on both the high and low kV series. The DEratio (referred to as ‘rel. Calcium’ in syngo.via) was then increased stepwise, and for each step, a three-material decomposition was done. For each DEratio, the mean difference between the fluid with bone and without bone for each of the three fluids was calculated, by measuring the mean CT numbers on the calculated VNCa image. The DEratio which yielded the lowest mean error was then chosen as the base for three-material decomposition in the rest of the experiments. Other syngo.via specific settings in the bone marrow oedema application were set to resolution 1, maximum 1500 HU and threshold -300 HU.
Measurement of VNCa numbers
The VNCa numbers were measured using a custom network in MeVisLab (version 2.8.2, MeVisLab GmbH, Bremen, Germany) by a single investigator. Using the high-resolution reconstruction, six volumes of interest (VOIs) were placed; one in each of the six compartments. Each VOI was placed so that it contained as much of the compartment as possible, while avoiding entrapped air and the irregular border between the fluid with bone and without bone. VOIs contained 221,603, 235,183, and 189,754 voxels in the three fluid compartments without bone and 113,279, 81,423, and 116,106 voxels in the three fluid compartments with bone.
VOIs at identical positions were automatically transferred and reused for all scan series. Mean HU values in the VNCa images in each of the six compartments were automatically measured within MeVisLab. Image noise was defined as the mean standard deviation of the VNCa measurements in the three compartments containing fluid and bone. Accuracy was defined as the mean error between fluid with and without bone as outlined in Eq. 1.
$$ \mathrm{Mean}\ \mathrm{error}=\frac{\mid {\mathrm{YMA}}_{\mathrm{bone}}-{\mathrm{YMA}}_{\mathrm{w}/\mathrm{o}\mathrm{bone}}\mid +\mid {\mathrm{RMA}}_{\mathrm{bone}}-{\mathrm{RMA}}_{\mathrm{w}/\mathrm{o}\mathrm{bone}}\mid +\mid {\mathrm{Water}}_{\mathrm{bone}}-{\mathrm{Water}}_{\mathrm{w}/\mathrm{o}\mathrm{bone}}\mid }{3} $$
(1)
Statistics
Data was analysed using freely available statistical software R (version 3.4.1; R foundation, Vienna, Austria). The correlation between image noise and pitch, rotation time, sharpness of the reconstruction kernel, spectral separation and dose was calculated using the Spearman’s rank correlation coefficient (rs). Since dose and spectral separation were measured multiple times, they were normalised so that when determining the correlation between dose and image noise and mean error, the tube-voltage combination of 70 kV/Sn150 kV was defined as 1, while for spectral separation, the image noise and mean error at 80 mGy was defined as 1. Since the three-material decomposition was based on a custom calibration, which minimised the mean error for each tube-voltage combination individually, the influence of the spectral separation on the mean error could not be determined in this experiment. A p value below 0.05 was considered to indicate a significant correlation, with no correction applied for multiple testing.