Differentiation of Hemorrhage from Iodine Using Spectral Detector CT: A Phantom Study

BACKGROUND AND PURPOSE: Conventional CT often cannot distinguish hemorrhage from iodine extravasation following reperfusion therapy for acute ischemic stroke. We investigated the potential of spectral detector CT in differentiating these lesions. MATERIALS AND METHODS: Centrifuged blood with increasing hematocrit (5%–85%) was used to model hemorrhage. Pure blood, blood-iodine mixtures (75/25, 50/50, and 25/75 ratios), and iodine solutions (0–14 mg I/mL) were scanned in a phantom with attenuation ranging from 12 to 75 HU on conventional imaging. Conventional and virtual noncontrast attenuation was compared and investigated for correlation with calculation of relative virtual noncontrast attenuation. Values for all investigated categories were compared using the Mann-Whitney U test. Sensitivity and specificity of virtual noncontrast, relative virtual noncontrast, conventional CT attenuation, and iodine quantification for hemorrhage detection were determined with receiver operating characteristic analysis. RESULTS: Conventional image attenuation was not significantly different among all samples containing blood (P > .05), while virtual noncontrast attenuation showed a significant decrease with a decreasing blood component (P < .01) in all blood-iodine mixtures. Relative virtual noncontrast values were significantly different among all investigated categories (P < .01), with correct hemorrhagic component size estimation for all categories within a 95% confidence interval. Areas under the curve for hemorrhage detection were 0.97, 0.87, 0.29, and 0.16 for virtual noncontrast, relative virtual noncontrast, conventional CT attenuation, and iodine quantification, respectively. A ≥10-HU virtual noncontrast, ≥20-HU virtual noncontrast, ≥40% relative virtual noncontrast, and combined ≥10-HU virtual noncontrast and ≥40% relative virtual noncontrast attenuation threshold had a sensitivity/specificity for detecting hemorrhage of 100%/23%, 89%/95%, 100%/82%, and 100%/100%, respectively. CONCLUSIONS: Spectral detector CT can accurately differentiate blood from iodinated contrast in a phantom setting.

M aterial with similar attenuation can be difficult to distinguish on conventional CT. A common clinical illustration of this problem is the differentiation of iodine from hemorrhage because both are hyperdense on conventional unenhanced CT. In patients with acute ischemic stroke, intra-arterial thrombolytic therapy has been shown to decrease morbidity and mortality. 1 However, it has also been reported that this increases the risk of intracranial hemorrhage (ICH), with reported frequencies of 10%-15% and mortality up to 83% for symptomatic ICH. [2][3][4][5][6][7] Because of the disruption of the blood-brain barrier, contrast extravasation can occur during the procedure in 30%-50% of cases, impairing the detection of or the differentiation from ICH due to the overlap in density. 5,[8][9][10][11] As of this writing, unenhanced conventional CT is performed within 24 hours after treatment for the detection of complications. In case of unclear findings (eg, if differentiation between ICH and iodine extravasation is not possible), follow-up imaging may be performed. Hence, the ability to differentiate these 2 is highly desirable to avoid additional examinations and to ensure appropriate management. 2 Two major discriminators of attenuation within a voxel are the energy of the x-ray beam and the concentration of attenuating material in that voxel. The presence or concentration of each discriminator in that voxel therefore cannot be determined by performing a single attenuation measurement from a broad photon energy spectrum. However, in dual-energy CT, attenuation at different energies is registered. In single-and dual-source dual-energy Revolution CT, this is achieved using a fast-kilovolt(peak) switching x-ray source (GE Healthcare, Milwaukee, Wisconsin) or 2 x-ray sources and 2 detectors (Somatom Force, Siemens Healthineers, Forchheim, Germany), respectively. Spectral detector CT (SDCT) (IQon; Philips Healthcare, Best, the Netherlands) distinguishes low-and high-energy data at the level of the detector using a dual-layer detector. The bottom and top layer absorb high-and low-energy photons, respectively. The main advantage of the latter approach is that spectral data are collected without the need to prospectively choose a spectral scanning protocol or mode, as needed in other dual-energy CT approaches.
Previous studies have shown the potential of dual-energy CT for differentiating iodine from hemorrhage in clinical settings 10,12-16 ; however, no study has been performed using SDCT for this application nor has the sensitivity or specificity been determined in a phantom system.
In our study, we investigated the ability of SDCT to differentiate ICH from iodinated contrast in a phantom model.

Image Acquisition
All scans were performed with a 464 Phantom (Gammex, Madison, Wisconsin) with a plastic 50-mL tube in its air cavity, filled with samples simulating hemorrhage (blood), diluted iodine, and a hemorrhage-iodine (blood-iodine) mixture. Each sample was scanned in the phantom with the following settings on the IQon SDCT scanner: 200 mAs, 120 kVp, pitch ϭ 1, gantry rotation time ϭ 330 ms, detector collimation ϭ 64 ϫ 0.625 mm, volume CT dose index ϭ 18 mGy. Conventional CT images were reconstructed using an iterative reconstruction algorithm (iDose 4, Level 3; Philips Healthcare), while the spectral images were reconstructed using a spectral reconstruction algorithm (Spectral B, level 0). Reconstructed slice thickness was 3 mm. All scans were performed 3 times to account for interscan variability.
Blood-iodine mixtures with matching densities were scanned with following proportions:

Image Analysis
All image analysis was performed using the proprietary image viewer (Spectral Diagnostic Suite; Philips Healthcare) of the vendor. Each diluted blood, diluted iodine, and blood-iodine mixture sample was analyzed by placing a circular 2-cm 2 ROI centrally in the inserts. The attenuation on the conventional and virtual noncontrast (VNC) image (using VNC image) and iodine concentration (using the iodine density map) were measured within each ROI (Fig 1). Relative VNC (R-VNC) attenuation (%) is calculated using the following equation:

Statistical Analysis
Statistical analysis was performed using SPSS 21.0 (IBM, Armonk, New York). Iodine-quantification measurements were compared with true iodine concentrations for correlation. Mean iodine quantification error (milligram/milliliter) was calculated and presented with a 95% confidence interval and Bland-Altman plot analysis. Attenuation in the conventional and VNC images was compared for correlation using the Pearson correlation. Attenuation on conventional images, VNC attenuation, and R-VNC are reported with 95% CIs for the investigated sample compositions (blood, [2 ⁄ 3]

RESULTS
Mean attenuation on the conventional and VNC images (Ϯ 95% CI) for the pure blood samples was similar: respectively, 39.1 Ϯ 31.5-46.7 HU and 39.1 Ϯ 31.5-46.7 HU (Fig 2 and the Table). Correlation between the hematocrit level in our samples and VNC attenuation was excellent (R 2 ϭ 0.97, P Ͻ .01), while correlation with the attenuation on the conventional images was moderate, but still significant (R 2 ϭ 0.50, P Ͻ .01) (Fig 3).
For the samples consisting only of diluted iodine, mean attenuation on conventional and VNC images (Ϯ 95% CI) was 129.5Ϯ 99.4 -159.6 HU and 14.5 Ϯ 13.4 -15.6 HU, respectively, which were both significantly different from other samples containing a blood component (P Ͻ .01) (Fig 2 and the Table). When one interprets these results, it is important to note that phosphate buffered saline has different attenuation properties from normal saline. The average attenuation of samples filled with only phosphate buffered saline was 8.5 and 8.3 HU on conventional and VNC images, respectively. Correlation between conventional and VNC attenuation for samples with pure iodine dilutions was not significant (R 2 ϭ 0.77, P ϭ .08).
The differences between R-VNC attenuation values of all compositions were significant (P Ͻ .01), as shown in Fig 4 and  There was a significant correlation between measured and true iodine concentrations (R 2 Ͼ 0.99, P Ͻ .01), with a mean iodine quantification error of Ϫ0.41 Ϯ 0.31-0.50 mg/mL) (Fig 5).
Receiver operating characteristic curve analysis for hemorrhagic-component detection showed a significant difference among areas under the curve (P Ͻ .01), being highest for VNC attenuation (0.97 Ϯ 0.94 -0.99 HU), followed by R-VNC attenuation (0.87 Ϯ 0.77-0.97 HU) and attenuation in the conventional CT images (0.29 Ϯ 0.16 -0.41 HU), and lowest for iodine quantification (0.16 Ϯ 0.06 -0.25 HU) ( Fig 6A); these differences were significant (P Ͻ .01). Using a threshold of Ն10 and Ն20 HU for VNC had a sensitivity/specificity of 100%/23% and 89%/95%, respectively. An R-VNC attenuation of Ն40% had a sensitivity/ specificity of 100%/82%. Conversely, using a threshold of Ն10 and Ն20 HU on the conventional CT images had a sensitivity/ specificity of 100%/13% and 94%/15%, respectively (Fig 6). Attenuation values (HU) of diluted blood, blood-iodine mixtures, and diluted iodine on spectral detector CT conventional and virtual noncontrast images. There is a significant incremental decrease of VNC attenuation values with decreasing blood content. The asterisk indicates a significant difference of conventional CT attenuation compared with other compositions (P Ͻ .01); double asterisks, significant differences of VNC attenuation among these compositions (P Ͻ .01).

DISCUSSION
Our results show that SDCT has excellent diagnostic accuracy for differentiating blood from iodinated contrast in a phantom. We observed similar sensitivity and specificity (both Ͼ90%) for hemorrhage detection using spectral dual-layer VNC images, compared with previous clinical studies that used single-or dualsource dual-energy CT. 10,13,15 These studies used visual assessment by radiologists for classifying hyperdensities on "simulated" conventional head CT images as hemorrhage or iodine; hyperdensities visible on VNC images were classified as having a hemorrhagic component. Although typical ICHs are hyperdense (Ͼ50 HU), they can be associated with a lower density due to anticoagulation, the presence of CSF with arachnoid laceration, or severe anemia (eg, sickle cell anemia), which can complicate subjective assessment. 20-29 Our results show high accuracy by quantitative VNC assessment for blood detection, including low densities on conventional images (as low as 10 HU). By combining a Ն10 HU VNC and Ն40% R-VNC cutoff, we observed a sensitivity and specificity of 100% for blood, including low densities and mixtures with iodine on conventional images (Fig 6). Several authors have investigated the accuracy of SDCT for material decomposition, which allows iodine to be subtracted from an image. These studies have shown high accuracy for iodinequantification and VNC attenuation values, though results vary. Our results show slightly lower iodine-quantification accuracy than recent publications, which can be explained by the extremely low iodine concentrations needed in our study, though results are still excellent. We had results comparable with those of Pelgrim et al, 30 which showed a median error of Ϫ0.6 mg/mL, while more recent studies showed differences ranging from Ϫ0.46 to 0.1 mg/mL. 31 Regarding VNC, Duan et al 31 showed good agreement between measured spectral detector VNC attenuation values and reference standards (Ϫ9.95-6.41 HU), confirmed by our study. Still, VNC inaccuracies can occur. When we incorporated a comparison of the attenuation between VNC and conventional images, small inaccuracies of true VNC values can be negated, which can explain the excellent results of R-VNC in our study for hemorrhage detection (100% sensitivity and 82% specificity) and size estimation (Figs 4 and 5).
Clinical studies using SDCT differentiating iodine from blood in skull imaging have been rare. As of this writing, the authors found only a pilot study from Cho et al, 32 showing that spectral data analysis can be helpful in discriminating ICH from contrast enhancement in intracranial malignancies. Regarding the detection of hemorrhage, a recent study of Nute et al 33 used singlesource dual-energy CT for distinguishing ICH from calcification in a phantom model with densities from 40 to 100 HU, resulting in an accuracy of Ͼ90%.
There are several limitations to our study. First, our results are without clinical data because this was not within the scope of our investigation. Our goal was to investigate in a phantom setting. Second, our hemorrhage dilutions were prepared with phosphate buffered saline to prevent red blood cell hemolysis, but this could potentially bias the attenuation characteristics of blood and iodine in our results: Phosphate buffered saline is not representative of brain tissue in a clinical setting. Third, our mixtures were homogeneously mixed for optimal spectral analysis, which may not always be the case in a clinical setting. Last, we used a relatively large 2-cm 2 ROI for our measurements, so our results could not be affected by spatial resolution and partial volume limitations. This might not always be feasible in patients if the size of the clinical ROI is small.

CONCLUSIONS
Our results show that the added spectral information of SDCT has high sensitivity and specificity in detecting blood and can accurately estimate hemorrhagic component size, including when mixed with iodine. This information could be of potential benefit in brain imaging for patients following reperfusion therapy in acute stroke.