Iodine accumulation of the liver in patients treated with amiodarone can be unmasked using material decomposition from multiphase spectral-detector CT.

Amiodarone accumulates in the liver, where it increases x-ray attenuation due to its iodine content. We evaluated liver attenuation in patients treated and not treated with amiodarone using true-non-contrast (TNC) and virtual-non-contrast (VNC) images acquired with spectral-detector-CT (SDCT). 142 patients, of which 21 have been treated with amiodarone, receiving SDCT-examinations (unenhanced-chest CT [TNC], CT-angiography of chest and abdomen [CTA-Chest, CTA-Abdomen]) were included. TNC, CTA-Chest, CTA-Abdomen, and corresponding VNC-images (VNC-Chest, VNC-Abdomen) were reconstructed. Liver-attenuation-index (LAI) was calculated as difference between liver- and spleen-attenuation. Liver-attenuation and LAI derived from TNC-images of patients receiving amiodarone were higher. Contrary to TNC, liver-attenuation and LAI were not higher in amiodarone patients in VNC-Chest and in VNC-Abdomen. To verify these initial results, a phantom scan was performed and an additional patient cohort included, both confirming that VNC is viable of accurately subtracting iodine of hepatic amiodarone-deposits. This might help to monitor liver-attenuation more accurately and thereby detect liver steatosis as a sign of liver damage earlier as well as to verify amiodarone accumulation in the liver.


Results
The 142 included patients comprised 58 women and 63 men with a mean age of 81.2 ± 8.6 years, ranging from 56-97 years. In total, five patients showed signs of severe liver steatosis indicated by a reduced liver attenuation (<40 HU) and/or reduced liver attenuation index (LAI, <−10 HU). The 21 patients treated with amiodarone received the fowling daily dosages: 100 mg/day, n = 1; 200 mg/day, n = 15, 400 mg/day, n = 5. Of these 21 patients 9 were woman, while 12 were men with a mean age of 77.0 ± 10.7 years, ranging from 56-99 years (for more detailed information see Table 1). There was no statistical difference in distribution between patients receiving amiodarone treatment compared to the patients not receiving amiodarone treatment regarding body mass index, history of smoking, and presence of diabetes, arterial hypertension, abnormal liver laboratory tests, as well as chronic liver disease. www.nature.com/scientificreports www.nature.com/scientificreports/ Objective assessment. The Shapiro-Wilk test did not show normal distribution for values in all assessed groups (p < 0.05). The 21 patients receiving amiodarone treatment showed significantly higher liver attenuation of 63.1 ± 9.8 HU on true non-contrast (TNC) images compared to 121 patients not receiving amiodarone treatment (58.1 ± 8.6 HU, p = 0.04, Table 2, Fig. 1A). See Fig. 2 as an example for elevated liver attenuation in TNC. Attenuation values in patients with amiodarone treatment ranged from 47.6 to 94.3 HU. Applying the LAI (intended to offer more robustness regarding patient specific differences in attenuation), the same difference could be observed, showing increased values of LAI in patients treated with amiodarone (16.8 ± 7.9 HU versus 9.7 ± 8.1 HU, p < 0.001, Table 3, Fig. 1A). LAI values for amiodarone treatment patients ranged from 5.0 to 38.5 HU. If all patients with definite signs of liver steatosis (n = 5) were removed liver attenuation and LAI in TNC images were still significantly higher in amiodarone patients (p < 0.05).
For evaluation of VNC accuracy in CTA-Chest-and -Abdomen, we excluded amiodarone patients to avoid influence of elevated liver attenuation from amiodarone accumulation in the liver. In liver and spleen, the 121 patients without amiodarone treatment showed VNC-Chest and -Abdomen attenuation values very similar to TNC (average liver attenuation: TNC, 58. In the additional 145 TAVR patients that also received VNC reconstructions of the TNC scans, the eleven patients receiving amiodarone treatment showed significantly higher liver attenuation of 61.6 ± 13.3 HU and LAI of 18.5 ± 8.7 on TNC compared to the 134 patients not receiving amiodarone treatment (liver attenuation, 54.4 ± 8.2 HU, p = 0.025; LAI, 10.1 ± 9.1 HU, p = 0.008, Supplementary Fig. 1). Contrary, in patients treated with amiodarone in VNC applied to the TNC images liver attenuation (52.9 ± 9.3 HU) and LAI (7.7 ± 8.3 HU) were comparable without significant differences and not higher compared to patients that have not been treated with amiodarone (liver attenuation, 53.9 ± 8.3 HU, p > 0.05; LAI, 9.7 ± 8.5 HU, p > 0.05, Supplementary Fig. 1).
In the phantom scan, attenuation of the minced liver without any additions was 28.9 ± 0.9 HU in TNC and 28.6 ± 1.0 HU in VNC. The minced liver mixed with amiodarone hydrochloride displayed attenuation values of 40.7 ± 1.1 HU in TNC images; in VNC attenuation was then reduced to values comparable to minced liver without any additions (31.9 ± 0.9 HU).  Table 2. Liver attenuation. Attenuation in HU, LAI -Liver attenuation index, TNC -true non-contrast, VNCvirtual non-contrast, significant differences in bold. Contrary to TNC, liver attenuation and LAI in VNC-Chest and -Abdomen images of patients treated with amiodarone were not higher compared to patients that have not been treated with amiodarone indicating that VNC not only subtracts iodine from contrast media but might also be able to subtract the iodine in amiodarone accumulated in the liver.

Discussion
This study quantitatively evaluated liver and spleen attenuation in patients with and without amiodarone treatment using TNC and VNC from spectral-detector CT. Patients that were being treated with amiodarone showed higher values of liver attenuation and LAI in TNC images. This is in line with previous studies 13,18,19 and case Attenuation of the liver is increased in TNC (~95 HU) which is likely to be caused by amiodarone treatment and accumulation in the liver. Liver attenuation is noticeably lower in VNC-Chest and -Abdomen (~54-57 HU) compared to TNC (~95 HU) indicating that VNC not only subtracts iodine from contrast media but might also be able to subtract iodine from amiodarone deposits. However, attenuation in the spleen (where amiodarone usually does not accumulate) is comparable between VNC Chest-and -Abdomen (~45 HU) and TNC (~56 HU). (B) Images of Fig. 3A Table 3. Liver attenuation index. Attenuation in HU, LAI -Liver attenuation index, TNC -true non-contrast, VNC -virtual non-contrast, significant differences in bold. www.nature.com/scientificreports www.nature.com/scientificreports/ reports 7,28,29 suggesting a connection between amiodarone treatment and elevated liver attenuation. Besides the aforementioned case reports, there is only one recent study examining the influence of amiodarone treatment on liver attenuation in CT. The study from Matsuda et al. showed an increase of mean liver attenuation of 9.7 HU in 25 patients after amiodarone treatment had been initiated. This is in line with our investigation that revealed also an increase of liver attenuation by 5.0 HU and an increase of LAI by 7.1 HU. Our results and earlier studies suggest that the parenchymal attenuation increase might be caused by the iodine bound to amiodarone molecule accumulated in the liver, similar to the way iodinated contrast media increases attenuation 7,13-16, [30][31][32] .
VNC-Chest and -Abdomen derived from spectral-detector CT in a cohort of TAVR planning patients not receiving amiodarone treatment showed accurate removal of iodine in both contrast phases (CTA-Chest and -Abdomen) with attenuation values very similar to TNC. In general, for quantitative liver assessment VNC might therefore serve as a replacement to unenhanced CT examinations when potential limitations of this technique are considered 33,34 . For example, Durieux et al. 2018 evaluated VNC images for abdominal depiction from a third generation dual-source dual-energy CT and pointed out that despite high accuracies of attenuation values for the liver and spleen with mean attenuation differences between 0-2 HU, there have been substantial differences in fluid, fat and renal tissue with mean differences as high as 32 HU. Other recent dual-energy studies also showed accurate iodine subtraction by VNC for depiction of the liver 23 Our study showed differences between TNC and VNC of 1.8 ± 4.6 HU for CTA-Chest and −1.1 ± 5.9 HU in CTA-Abdomen in liver. These smaller differences in accuracy between the dual-energy CT solutions available might to some extend also be explained by their technical approaches. The spectral-detector CT, for example, separates high and low energy photons on the detector level so that acquired data are exactly matched temporally and spatially. Thereby material decomposition is enabled in the raw data domain 21,23,37,38 .
Dual-energy CT allows for identification of iodine containing voxels. In a second step, the iodine component of a predefined voxel can be measured and accurately subtracted to create VNC images. This approach should be equally valid for iodinated contrast media and iodine bound to amiodarone 21 which have similar molecular structures. The iodine molecules are connected to a benzene ring and iodine atoms have a comparable share of the total molecular mass 9-16 . In our study, contrary to TNC, liver attenuation and LAI in VNC of CTA-Chest and -Abdomen examinations did not differ significantly between patients treated and not treated with amiodarone, suggesting that VNC also subtracts iodine from iodine-containing amiodarone accumulation in the liver. These initial findings were then verified in a phantom scan, in which VNC of TNC were able to subtract attenuation of iodine-containing amiodarone. Further, the results from our first analysis could be confirmed in an additional patient cohort, for which, unlike to our initial patient cohort, also VNC reconstructions of TNC images were available. In line with our initial results VNC was also able to reduce liver attenuation and LAI in patients treated with amiodarone to a level comparable to patients not treated with amiodarone.
Our patient cohort only included five patients with reduced liver attenuation and LAI indicating liver steatosis, therefore we could not add a further pathology focused analysis on liver steatosis. Still, our data suggests that VNC could be relevant in these patients. In the group of amiodarone patients, VNC led to a mean decrease of LAI of around 10 HU compared to TNC, potentially representing artificially high attenuation in TNC from the iodine in amiodarone. This difference can be relevant in amiodarone patients when it obscures potential liver damage reflected by liver steatosis 1,7,8 . The LAI is predictive for a liver fat fraction of >30% and image-based diagnosis of liver steatosis when the difference is less than −10 HU 39-41 , hence a mean difference of 10 HU could possibly lead to many misclassified patients. Due to its toxicity amiodarone is also known to independently induce steatohepatitis 30 which can then result in liver cirrhosis, drastically increasing risk of mortality 1,7,8 . Further, liver steatosis, that is caused for other reasons than amiodarone, is common in a general population with non-alcoholic liver steatosis as its most common subtype and an incidence of 15% in general population 24,42 . A severe liver fat fraction of ≥30% is considered to have relevant clinical implications 43,44 but could be obscured in amiodarone patients due to amiodarone accumulation in liver 24,45,46 . As described above, VNC can also overestimate attenuation of fat; therefore, it needs to be considered that VNC can negatively affect CT attenuation-based diagnosis of liver steatosis in CT imaging 24 .
For patients with increased liver attenuation undergoing amiodarone treatment, material decomposition and TNC might give radiologists and clinicians more confidence to identify amiodarone as the cause of increased liver attenuation 7,28,29 , potentially allowing for more precise monitoring of liver attenuation and earlier adaption of drug treatment. Identification and quantification of amiodarone in the liver might also be useful as additional guidance for amiodarone treatment, as tissue levels have been suggested to be useful to anticipate toxic liver damage and treatment efficacy 7,18,29 . Differences in liver attenuation between TNC and VNC could be used as a measure to quantify amiodarone accumulation in the liver 21,47 . Thus, spectral imaging might serve as a quantitative biomarker for treatment monitoring in patients before and/or under amiodarone with and with-out fatty liver disease, which should be evaluated in future studies.
The study has some limitations that need to be considered. Spectral imaging data for the unenhanced chest acquisition (TNC) is not routinely saved at our institution, therefore creation of VNC images from this phase was not possible in the initial cohort. However, to address this shortcoming we changed our image reconstruction protocol and included an additional cohort in which VNC was also reconstructed from TNC scans; however, we refrained from conducting a pooled analysis as in the initial patient group VNC was performed on contrast enhanced CT scans and in the additional patient group VNC was performed on unenhanced scans. In this cohort we were able to show that VNC of TNC can reliably reduce increased liver density due to amiodarone deposition. As frequently applied in comparable studies, we used ROI-based measurements of mean attenuation 22 www.nature.com/scientificreports www.nature.com/scientificreports/ a volumetric assessment of attenuation covering an entire organ might be less susceptible for any measuring bias. To minimize mismeasurement, we used two ROI for spleen and four ROI for the liver with a standardized size and averaged the results. Further, we placed ROIs in the same anatomical locations in each image set with the help of screenshots of the initial placement.
In conclusion, patients treated with amiodarone showed significantly higher liver attenuation in TNC. In patients not treated with amiodarone VNC accurately subtracted contrast-agent related iodine in contrast-enhanced examinations. Also, VNC depicted comparable liver attenuation in patients treated and untreated with amiodarone indicating that VNC subtracted iodine resulting contrast injection as well as from hepatic amiodarone deposits.

Materials and Methods
The institutional-review-board approved this retrospectively designed study. The standards of the Health-Insurance-Portability-and Accountability-Act were followed. Under Code-of-Federal-Regulations (title 45, §46.116d) informed written consent was waived. Following criteria for patient inclusion were applied: (i) patient age ≥18 years, (ii) patients with severe aortic stenosis receiving a TAVR planning examinations on a clinical spectral-detector CT system (IQon, Philips Healthcare) between May 2017 and September 2018, (iii) complete imaging data set, including the following examinations in the given order, unenhanced chest [TNC], CT angiography chest [CTA-Chest], and CTA abdomen [CTA-Abdomen]), (iv) at least two thirds of the spleen as well of the liver depicted in all three acquisitions, (v) sufficient data available on drug and/or amiodarone treatment, respectively, and (v) no reports of multiple blood transfusions, hemochromatosis, Wilson's and/or glycogen storage diseases in patient history to exclude possible confounder for elevated liver attenuation 48 , (Fig. 3). TNC chest examinations were executed to perform aortic valve calcium scoring prior to TAVR procedure. This resulted in total of 142 patients included in the final analysis. Patients that were treated with amiodarone at time of the TAVR examination were grouped in the amiodarone cohort (n = 21), patients without amiodarone treatment were grouped in the no amiodarone treatment cohort (n = 121). Body mass index, history of smoking, and presence of diabetes, arterial hypertension, abnormal liver laboratory tests, as well as chronic liver disease were recorded for each patient to detect any potential confounders between the two groups. Table 4 for the unenhanced chest scan, CTA-Chest and CTA-Abdomen. For CTA-Chest examinations, 25 ml of iodinated contrast agent (Optiray 350, Guerbet) were administered through an antecubital vein. A retrospective ECG-gated helical scan was initiated using bolus-tracking: a region-of-interest (ROI) was placed in the left atrium and scans were initiated after an attenuation threshold of 60 Hounsfield units (HU) was reached. For CTA-Abdomen, another 25 ml of contrast agent were administered 120 seconds after the chest scan. The scan was started using bolus-tracking with a ROI in the descending aorta when attenuation increased 10 HU over the baseline. This threshold was chosen, as contrast in vessels from the previous scan was still present but differing between patients due to individual differences in circulation. After testing we therefore decided to use an increase of 10 HU over the baseline instead of an absolute threshold. Iterative reconstructions were performed for unenhanced chest scan (TNC), CTA-Chest and CTA-Abdomen. VNC were created from CTA-Chest and CTA-Abdomen data using the vendor's dedicated material decomposition and spectral reconstruction algorithms (Spectral B, level 3, Philips Healthcare).

Additional patient cohort.
To verify if VNC can truly remove attenuation from amiodarone accumulation of the liver we included additional TAVR patients (n = 145) receiving the protocol from above and outlined in Table 4 between December 2018 and June 2019. For these patients, reconstructions of VNC images from the TNC scans were available. Eleven of these patients received amiodarone treatment (supplementary Table 1).  www.nature.com/scientificreports www.nature.com/scientificreports/ Objective analysis. Circular ROIs of at least 500 mm 2 were placed on axial images to measure the mean attenuation in HU by a radiologist with four years of experience in abdominal imaging. ROI were first drawn in CTA-Abdomen (because anatomical identification and delineation of structures was most accurate here) and subsequently were transferred to TNC and CTA-Chest as well as VNC-Chest and -Abdomen using the vendor's proprietary image viewer (IntelliSpace Portal v9, Philips Healthcare). ROI positions could differ for image sets derived from different acquisitions due to respiratory motion. To minimize any mismeasurements from different ROI locations, screenshots of each ROI in CTA-Abdomen images were used to match placement in the other image sets. Two ROIs each were placed in the right and left liver lobe as well as in the spleen. The results of the four ROIs from the right and left liver lobes were averaged to obtain results for the entire liver. The two ROIs of the spleen were also averaged. The liver attenuation index was calculated as the difference of attenuation between liver and spleen 13,41 . Phantom measurements. In addition, we also performed a phantom analysis using falcon tubes placed in a non-anthropomorphic water phantom using a helical scan with the scan parameters from the unenhanced scan of the chest. The falcon tubes were filled with 30 g of minced pig's liver; once without any additions and once with amiodarone hydrochloride aiming for a concentration of 0.2 mg I/ml 49 . The water phantom with the falcon tubes were scanned three times each. Three ROIs on different levels were placed centered in each of the two falcon tubes for each of the three scans and then averaged.
Statistical analysis. Statistical analysis was conducted using JMP Software (V14, SAS Institute).
Quantitative results are presented as mean ± SD. Shapiro-Wilk test was used to test for normal distribution. Wilcoxon rank-sum test was used to test for any difference of attenuation in between patients treated and not treated with amiodarone. Mann-Whitney-U test was used to test for any difference of attenuation between TNC and VNC. Statistical significance was defined as p < 0.05. To estimate number of patients to evaluate if VNC can subtract attenuation of iodine from amiodarone we conducted a power analysis, using G*Power (v. 3.1.9.4) 50 . According to the literature, the mean attenuation of liver was assumed to be 60 HU 51 with a standard deviation of approximately 6 HU 52 . We estimated that mean accumulation of amiodarone in the liver should result in an increase of attenuation around 5 HU, resulting in an estimated effect size of 0.83. We aimed for a power of at least 0.8. Assuming a distribution between groups of 1:6 (amiodarone vs. no amiodarone treatment), our sample size should be at least 98 patients (84 patients no amiodarone, 14 patients with amiodarone treatment).

Data availability
The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.