Feasibility and accuracy of dual-layer spectral detector computed tomography for quantification of gadolinium: a phantom study

An anthropomorphic chest phantom (QRM GmbH, Moehrendorf, Germany) was used. The phantom resembles a chest with corresponding X-ray attenuation behaviour. The phantom has a cylindrical cardiac chamber in which a plastic holder was placed (Fig. 1). Three plastic holders were made, two consisting of five tubular inserts, and one consisting of three tubular inserts with surrounding 2% agar gel solution. In addition, a plastic holder with one tubular insert containing water with surrounding 2% agar gel solution served as control. The fourteen 32-mL tubular inserts contained different concentrations of the gadolinium-based contrast agent gadobutrol (Gadovist 1.0, Bayer Healthcare, Berlin, Germany). One millilitre of this contrast agent contains 157.25 mg gadolinium. Different amounts of gadobutrol were diluted in water, resulting in the following concentrations of gadolinium: 0.0, 0.1, 0.2, 0.4, 0.5, 1.0, 2.0, 3.0, 4.0, 5.1, 10.6, 15.7, 20.7 and 26.3 mg/mL, which is equivalent to 0.000, 0.001, 0.002, 0.002, 0.003, 0.006, 0.013, 0.019, 0.026, 0.032, 0.068, 0.100, 0.132 and 0.167 mmol/mL, respectively.

Concentrations were chosen to mimic an estimated clinical range of gadolinium concentrations encountered after injection of 0.1–0.2 mmol of gadolinium per kilogram. Strich et al. [32] measured percentage dose of gadolinium-based contrast agent per gram of tissue in healthy rabbit organs 5 min after admission, resulting in the following percentages: 0.052%/g heart, 0.073%/g lungs, 0.037%/g liver, 0.037%/g spleen and 0.250%/g kidney. On the basis of these percentages, an estimation of gadolinium concentrations encountered at each organ can be calculated. At 31.5 mg/kg bodyweight (equal to 0.2 mmol/kg) gadolinium administration, a human subject of 70 kg would receive a total of 2201.5 mg gadolinium. On the basis of the percentages determined by Strich and colleagues, gadolinium distribution in the heart 5 min after injection would be 0.00052 × 2201.5 mg, or 1.15 mg per gram myocardium. Myocardial muscle has a specific gravity of 1.05 g/mL [33], implicating an estimated gadolinium concentration encountered in the myocardium of 1.15 mg/g × 1.05 g/mL, or 1.21 mg/mL. Using the aforementioned distribution percentages these calculations can also be applied to the lungs, liver, spleen and kidney, with a calculated estimated specific gravity (weight/volume) of 1.34, 1.01, 0.71 and 0.85 g/mL, respectively [34‐36]. Thus, it is to be expected that gadolinium concentrations of 2.15, 0.82, 0.58 and 4.67 mg/mL are encountered in healthy lung, liver, spleen and kidney tissue, respectively. These concentrations are in the range of concentrations evaluated in this study. As it is expected that in tissue with a perfusion defect lower concentrations of gadolinium will be encountered, we also evaluated ultra-low concentrations of gadolinium down to 0.1 mg/mL.

Images were acquired using the newest generation 64-detector row SDCT system (iQon Spectral CT, Philips Healthcare, Best, the Netherlands). This system uses a single X-ray tube and a dual-layer detector. The detector separates the X-ray beam into low (upper layer) and high (lower layer) energy data, which is used to reconstruct spectral-based images (SBI). The SBI contain the raw data of both layers and are used to reconstruct any dual-energy image and/or analysis. In addition, by combining the output of both layers, a conventional image is reconstructed from the data. The phantom was imaged in spiral mode at 120 and 140 kVp. The tube current–time product was set to a fixed value of 200 mAs, resulting in a volumetric CT dose index (CTDI_vol) of 18.4 and 26.5 mGy for 120 and 140 kVp acquisitions, respectively. The following parameters were used: detector collimation 64 × 0.625 mm, rotation time 0.4 s and pitch 1.046. At both tube voltages, acquisitions were repeated five times with small displacements between each acquisition to take into account interscan variation. Thus, the phantom was translated a few millimetres in the left–right direction, as well as along the z-axis of the CT scanner. After the five repetitions, the phantom was set back to the starting position.

The raw projection data from both detector layers were automatically reconstructed into SBI. Subsequently, MDI was performed in the projection domain, which efficiently eliminates beam hardening artefacts [37]. All images were reconstructed with standard chest reconstruction filter B and spectral level 3. Spectral is a model-based iterative reconstruction developed for the SDCT, it is an equivalent to iterative model-based reconstruction (IMR). Spectral consists of six levels, whereby a higher spectral level implies more noise reduction. Slice thickness and increment were both 1 mm. The reconstructed images were evaluated on a dedicated workstation using the Spectral CT Viewer (IntelliSpace Portal v6.5.0.02080, Philips Healthcare, Best, the Netherlands).

On three different slices of each data set a region of interest (ROI) with a fixed size of 225 mm² was drawn in the centre of each insert (Fig. 2a). Subsequently spectral plots of every ROI were obtained, in which mean Hounsfield units (HU) were plotted as a function of different energy levels expressed in kilo electron volt (keV) (Fig. 2b). These mean HU values of the spectral plots were extracted in steps of 10 keV and used as an input for the analysis. The currently used SDCT system uses traditional integrated detectors at two energy spectra and is therefore not able to image and/or quantify a material-specific K-edge [37]. Materials with a K-edge within the SDCT range (40–200 keV) will not show a discontinuity in their attenuation function on the SDCT spectral plot. When evaluating the mean attenuation across monochromatic energies, this does not pose a problem and therefore the whole energy spectrum can be used (40–200 keV). However, for the quantitative analyses of gadolinium concentrations a comparison is made with the attenuation profile of pure gadolinium which does contain the discontinuity in their attenuation function at the K-edge. Therefore, to take into account the non-linear energy dependency close to the K-edge of gadolinium (50.2 keV), only the energy range from 70 to 200 keV was used for the quantitative analysis. With in-house-developed software, attenuation profiles were reconstructed from the provided mean HU, and gadolinium concentrations were calculated by fitting combinations of known attenuation profiles of pure gadolinium and water to the reconstructed attenuation profiles (Fig. 2c). For each ROI drawn in the phantom, the in-house-developed software assumed that all voxels within this ROI were composed of only gadolinium and water and that the sum of these fractions added up to 100%. Known attenuation profiles of pure gadolinium and water were obtained from the National Institute of Standards and Technology (NIST) database [38]. Therefore, no calibration scans with water and/or gadolinium concentrations were needed. For all thirteen different gadolinium concentrations, 15 measurements were performed at both 120 and 140 kVp (three slices, five repetitions). In addition, 30 measurements (15 at both 120 and 140 kVp) were performed on the control phantom. Gadolinium concentrations were calculated for each measurement.

CT attenuation during injection of low gadolinium concentrations (i.e. 0.1–0.2 mmol/kg bodyweight) will generally lead to lower HU values compared to the use of iodinated contrast agents [24, 25] . To investigate the ability of SDCT to visually identify an increase in HU values due to the presence of a gadolinium-containing contrast agent we extracted mean attenuation coefficients across monochromatic energies (40–200 keV) for the different gadolinium concentrations used in this study (Fig. 3).

To evaluate the quantification accuracy of gadolinium concentrations, we defined measurement errors in milligrams per millilitre and relative measurement errors in percentages. Measurement errors were calculated by subtracting true gadolinium concentrations from the measured gadolinium concentrations. Subsequently, relative measurement errors (%) were calculated as follows:

All measurement error analyses were performed separately for 120 and 140 kVp. In addition, sub-analyses were done for each concentration. The Shapiro–Wilk test was used to identify normally distributed data. For each concentration, statistical differences of measurement errors between 120 and 140 kVp were analysed using paired t test for normally distributed data. A Bonferroni corrected P < 0.004 (0.05/number of comparisons) was considered significant. Pearson’s correlation coefficient was used to evaluate correlations between measured and true gadolinium concentrations at different tube voltages and for each scan repetition. In addition, reproducibility was evaluated. To define agreement of results, the two-way random single measure intraclass correlation coefficient (ICC) with corresponding confidence interval (CI) was used for all possible two-way interactions. ICCs between 0.61 and 0.80 were considered good and ICCs greater than 0.80 excellent [39]. Measurement interscan variabilities of all scan repetitions were plotted in one single plot by using a modified Bland–Altman plot described by Jones et al. [40]. In this figure the measurement differences of every measurement compared to the mean measurement of all scans are plotted against the mean measurement of all scans. As described by Jones et al., the limits of agreement were calculated as mean ± 1.96 × SD, where the SD is an estimate of the standard deviation for all observers [40]. Values are listed as mean with standard deviation (SD), unless stated otherwise. A P value less than 0.05 was used to indicate statistical significance. IBM SPSS version 21.0 (IBM corp., Armonk, New York, USA) was used for statistical analyses.