Material decomposition using iodine quantification on spectral CT for characterising nodules in the cirrhotic liver: a retrospective study

Background There is limited scientific evidence on the potential of spectral computed tomography (SCT) for differentiation of nodules in the cirrhotic liver. We aimed to assess SCT-generated material density (MD) parameters for nodule characterisation in cirrhosis. Methods Dynamic dual-energy SCT scans of cirrhotic patients performed over 3 years were retrospectively reviewed. They were classified as hepatocellular carcinoma (HCC), regenerative or indeterminate, according to the European Association for the Study of the Liver criteria. MD maps were generated to calculate the area under the curve (AUC) and cutoff values to discriminate these nodules in the hepatic arterial phase (HAP) and portal venous phase (PVP). MD maps included iodine concentration density (ICD) of the liver and nodule, lesion-to-normal liver ICD ratio (LNR) and difference in nodule ICD between HAP and PVP. Results Three hundred thirty nodules belonging to 300 patients (age 53.0 ± 12.7 years, mean ± standard deviation) were analysed at SCT (size 2.3 ± 0.8 cm, mean ± SD). One hundred thirty-three (40.3%) nodules were classified as HCC, 147 (44.5%) as regenerative and 50 (15.2%) as indeterminate. On histopathology, 136 (41.2%) nodules were classified as HCC, 183 (55.5%) as regenerative and 11 (3.3%) as dysplastic. All MD parameters on HAP and the nodule difference in ICD could discriminate pathologically proven HCC or potentially malignant nodules from regenerative nodules (p < 0.001). The AUC was 82.4% with a cutoff > 15.5 mg/mL for nodule ICD, 81.3% > 1.8 for LNR-HAP and 81.3% for difference in ICD > 3.5 mg/mL. Conclusion SCT-generated MD parameters are viable diagnostic tools for differentiating malignant or potentially malignant from benign nodules in the cirrhotic liver. Supplementary Information The online version contains supplementary material available at 10.1186/s41747-021-00220-6.

The main radiological system to classify cirrhotic nodules at CT and MRI is the Liver Imaging Reporting and Data System (LI-RADS) [14]. According to the LI-RADS, all cirrhotic nodules have the potential to be HCC, ranging from least likely, namely LI-RADS category 1 (definitely benign), to most, LI-RADS category 5 (definitely malignant) [14]. Mid-spectrum categories such as LI-RADS 3 or 4 fall into the indeterminate zone of diagnostic suspicion, i.e., they do not represent definite HCC nor benign regenerative nodules [15]. Such lesions, especially nodules between 1 and 2 cm in size, are often found to be dysplastic or well-differentiated HCC on pathological examination [16][17][18][19]. The most recent LI-RADS-based guidelines have shown a sensitivity of 89.6%, a specificity of 81.2% and an overall diagnostic accuracy of 88.0% for HCC lesions > 2 cm [20].
The advent of spectral computed tomography (SCT) and its ability to generate synchronous datasets of polychromatic and monochromatic images with material density (MD) maps has enabled radiologists to experiment beyond the conventional enhancement patternbased algorithms, which visually assess contrast uptake, retention and washout [21][22][23][24][25]. The most recent upgrades in CT hardware and software allowed the use of SCT for tissue characterisation using MD maps [26]. In fact, at SCT, images from two distinct x-ray spectra at 70-80 kVp (low energy) and 120-140 kVp (high energy) can be exploited to visualise hypervascular lesions with better conspicuity on MD datasets [25,27,28]. MD images are used to calculate the iodine concentration density (ICD) of the lesion, measured in milligrams per milliliter along with other parameters [29][30][31][32].
Most SCT liver studies have investigated the diagnosis and differentiation of HCC from benign hypervascular lesions such as haemangioma, fibronodular hyperplasia or angiomyolipoma in noncirrhotic populations [33][34][35][36]. However, there is a lack in existing scientific knowledge regarding the potential of SCT for the overall differentiation of nodules in the cirrhotic liver [37][38][39].
Thus, the aim of this study was to assess the diagnostic performance of SCT for the discrimination of malignant (HCC), potentially malignant (indeterminate) and benign (regenerative) nodules in cirrhosis patients.

Population
This single-centre retrospective study was approved by the institutional ethics board, and the requirement for informed consent was waived. We screened SCT scans of 380 cirrhotic patients performed from January 2016 to June 2019. SCT scans were obtained by searching the hospital and radiology database for the cirrhotic population who underwent pre-transplant and surveillance CT scans. For inclusion in the study, patients who underwent pre-transplant CT, as per institute protocol (n = 300), and those who were found to have liver nodules on CT, who subsequently underwent biopsy (n = 20), were included. Patients with identifiable nodules on SCT who did not undergo liver transplantation or follow-up with biopsy (n = 80) were excluded.
A total of 330 nodules from 300 selected SCT studies were independently evaluated by two observers with more than 4 years of experience in liver imaging. Conflicting observations between the two reviewers were reevaluated at a consensus meeting, and a final decision was taken in the presence of a third reviewer. The workflow of the study is shown in Fig. 1.

SCT protocol
Scans that were performed as a triple-phase study on the single-source, dual-energy, 64-slice CT scanner (Discovery CT 750 HD, General Electric Healthcare, USA) with fast kVp switching between 80 and 140 kVp were included. Scanning protocol and technique are summarised in Table 1.

Image analysis
The polychromatic image sets with 120 kVp photon energy, which are analogous to the conventional CT images, were used to classify cirrhotic nodules into HCC, indeterminate and regenerative based on enhancement pattern of lesions on hepatic arterial phase (HAP) and portal venous phase (PVP) (Fig. 2). Nodules with HAP hypervascularity and subsequent washout on PVP or delayed phase were classified as HCC [40]. Nodules that did not show classical features of HCC (neither hypervascular nor displaying washout) were classified as 'regenerative' (non-HCC by EASL criterion) (Fig. 3) [17]. Since we excluded all other liver lesions at the beginning of the study, benign lesions were synonymous for regenerative nodules for this study. Nodules with ambiguous enhancement, such as isolated arterial hypervascularity or exhibition of washout or capsule formation on PVP without arterial enhancement, were grouped as 'indeterminate' nodules [41].
The MD images generated by SCT were viewed and analysed with material decomposition and projection-based software on the workstation (CTAW software 4.4, General Electric Healthcare, USA), with a standard softtissue display window (preset at 40 HU and width 400 HU). Images were reviewed at a standard reconstruction kernel on HAP and PVP.
Iodine density of the nodule (ICD nodule ), liver parenchyma (ICD liver ) and aorta (ICD aorta ) were measured in milligrams per milliliter on HAP and PVP with the help of three discrete circular regions of interest (ROIs) of equal area, size and shape, carefully placed in the maximum enhancing portion of each structure, at the same level in the abdomen, as shown in Fig. 2. The average number of pixels in each ROI was 400 (range 310-560). Three parameters were calculated using the above values: lesion-to-normal liver ratio (LNR = ICD nodule /  ICD liver ), normalised iodine concentration (NIC = ICDnodule /ICD aorta ) and difference in ICD (ΔICD = ICD nodule HAP − ICD nodule PVP). The NIC was measured as a baseline reference for optimum enhancement, to minimise variations among patients.
For MD parameter analysis, nodules were organised into two broad groups: malignant (HCC), also including potentially malignant (indeterminate) nodules, versus benign (regenerative) nodules owing to the limitation of pathological confirmation of dysplastic nodules and the significance of their stepwise progression to frank HCC [8,10,11] (Fig. 4). Serum alpha-fetoprotein (AFP) values were recorded for all patients at the time the scan was performed, with observer blinding and used for statistical and comparative data analysis. Histopathology results of explant liver specimens (310 nodules) and nodule biopsy (20 nodules) were used as a reference standard.

Histopathological analysis
Percutaneous ultrasound-guided liver biopsy was performed using an 18-gauge cutting needle. Explant and biopsy specimens were routinely processed for systematic histological evaluation. All biopsies were read by a liver pathologist with 15 years of experience who was blinded to imaging results.
Regenerative nodules were classified on basis of features of a hyperplastic or a hepatocellular nodule at low magnification, without distinct cytoarchitectural features from the extralesional cirrhotic background, intralesional 2 cell-thick hepatocyte plates and unconnected lesion portal tracts [8]. Dysplastic nodules were characterised by architectural abnormalities such as 2 or 3 cell-thick hepatocyte plates, low-grade cytological atypia, nuclear crowding, inconstant detection of map-like clonal growth or nodule-in-nodule appearance, and inconstant increase in the number of unpaired arterioles which was confirmed by smooth muscle actin immunostaining. Progressive sinusoidal capillarisation was detected by CD34 immunostaining. Distinction from welldifferentiated HCC is known to be challenging [9- 11,16]. A diagnosis of HCC was finalised on seeing architectural abnormalities such as nuclear crowding, three or more cell-thick hepatocyte plates, relative cell monomorphism, presence of multiple microacinar structures with or without bile plugs, an abnormally high number of muscularised unpaired arterioles, capillarised vessels and infiltration of the portal tract, fibrous septa, or veins by single hepatocytes [9- 11,16].

Statistical analysis
Continuous data were presented as mean ± standard deviation (SD) or median with interquartile range (IQR) according to data normality. The comparison was made using a one-way analysis of variance (ANOVA) test, followed by a post hoc test by the Bonferroni method. Categorical data were reported as numbers and frequencies (%) and compared by using the χ 2 test or Fisher exact test as applicable. The agreement between two observers was assessed using kappa (κ) statistics. The comparison between malignant or potentially malignant (HCC and indeterminate) and benign (regenerative) nodules for comparing and analysing the significance of MD parameters on SCT was computed using the Mann-Whitney U test or independent t-test as appropriate. Sensitivity, specificity, positive predictive values and negative predictive values were calculated with their 95% confidence intervals (CI), according to the binomial distribution. The area under the curve (AUC) at receiver operating characteristic (ROC) analysis was also calculated. The quantification of effect (odds ratio, OR) was based on the nominal logistic regression where imaging discrimination of more than two nodules, i.e., HCC, indeterminate/dysplastic or regenerative, was calculated. The logistic regression analysis was used to quantify the effect and to derive cutoff values for malignant or potentially malignant and benign nodules. All analyses were performed using SPSS statistical software (version 22, IBM, USA). A p value less than 0.05 was considered to indicate statistical significance.

Results
The mean size of 330 analysed nodules was 2.3 ± 0.8 cm. Such 330 nodules were found in 300 cirrhotic patients of which 280 were pre-transplant candidates and 20 were under surveillance. The majority of patients (n = 266; 89%) were men; mean patients' age was 53.2 ± 12.7 years. The interobserver agreement between the two radiologists gave a κ value of 0.95. All details of imaging diagnosis and pathology correlation are shown in Table 2 and Supplementary Table 3. Enhancement characteristics of nodules are detailed in Table 3.
The MD parameters on PVP demonstrated AUC between 40 and 50%, sensitivity (50-58%) and specificity (50-60%) as depicted in Fig. 5 along with the cutoff values (Supplementary Table 4). MD parameters were analysed in both phases to define threshold values for differentiation of three cirrhotic nodules (Supplementary Figure 2) since pathology could not be considered as an absolute ground truth in case of dysplastic or regenerative nodules.

Discussion
The main finding from our study is that MD parameters such as ICD nodule HAP, LNR-HAP, NIC-HAP, ICD nodule PVP, NIC-PVP, and ΔICD provided adequate differentiation between the benign and malignant (including potentially malignant) nodules (all p < 0.05). This finding established the feasibility of SCT for diagnosis and classification of cirrhotic nodules. Our results depicted an excellent agreement between the expected enhancement and perfusion pattern of malignant and benign nodules in the cirrhotic liver. We also observed that few MD parameters (ICD nodule HAP, LNR-HAP, ΔICD, and NIC-HAP) showed higher diagnostic accuracy (all AUCs > 80%) than those previously reported (70%) by nonspectral CT [42]. MD parameters on PVP were not as accurate as on HAP. We observed that LNR-PVP did not demonstrate significant demarcation between nodules. This could possibly be due to the unremarkable difference in contrast density of nodules as compared to the underlying liver on PVP, vis-a-vis the contrast divergence usually observed on HAP. Although this observation needs further validation, it implies that more weightage should be given to MD parameters on HAP than on PVP.
Pfeiffer et al. [37] have demonstrated the efficiency of LNR in both HAP and PVP as a better predictor than low (65) keV SCT or MRI for characterising HCC, and Gao et al. [43] demonstrated that LNR-HAP was the best parameter on MD maps to differentiate HCC from other liver nodules in cirrhotic patients.
Our analysis highlighted that ICD nodule HAP is one of the best objective predictors for nodule differentiation.
However, ΔICD appears to be the single most dependable marker, to discriminate cirrhotic nodules. This is likely since it incorporates data from both HAP and PVP, yields high diagnostic accuracy (AUC 81%) and demonstrates distinct cutoff values for malignant versus benign nodules.
The unique capability of SCT to produce MD-based images which can be used to determine the composition of soft tissues has been extensively explored for oncological imaging [25,[44][45][46][47]. Matsui et. al [48] reported that liver nodules differentiate into cancer either via a de novo or a stepwise process where the intranodular hepatic vessels are gradually replaced by tumoural neovasculature. This change in nodule vascularity according to its morphological progression can be quantified by measuring their iodine absorption (K-edge 33.2 keV of iodine makes it ideal for decomposition) which is an indirect measure of their vascularity [49]. Up to date, there are no clear-cut criteria (from EASL or LI-RADS) for the definition of dysplastic or regenerative nodules [40]. However, it is well known from previous studies that dysplastic and regenerative nodules are distinct entities that are visualised in the cirrhotic liver on different modalities [50]. MD values, which are primarily based on tissue iodine density, are exclusive to the inquired lesion and are independent of observer reading or interpretation. They have the potential to serve as non-invasive biomarkers for liver nodules in cirrhosis.
Cho et. al [51] showed that hepatocarcinogenesis is a time-dependant progression of benign to malignant nodules and this was illustrated in our study, by the presence of regenerative nodules in younger patients with more HCC nodules observed in the older population. Rizal et al. [52] demonstrated that cirrhosis, regardless of aetiology, is more prevalent in men, worldwide. This gender predilection was seen involving all three groups in our study, which was in line with these observations. The median nodule size and AFP also showed a significant increase from the regenerative to the malignant group. This trend indicated a concomitant increment in lesion size and AFP with evolution to malignant disease [53].
The majority of HCC's on CT were correctly classified in comparison to pathology. Only 3% of the lesions, which demonstrated an atypical enhancement pattern, were misdiagnosed on imaging. These lesions measured < 1 cm on size, which was probably the reason for ambiguous enhancement.
Among the indeterminate nodules categorised on CT, only 10% were confirmed as dysplastic nodules on biopsy. We believe that this was because pathology could not adequately distinguish these nodules distinctly, as shown by previous studies, because low-grade dysplastic nodules either appear normal or show only minimal nuclear atypia and high-grade dysplastic nodules are extremely difficult to distinguish from well-differentiated HCC [9,10]. We believe that the rest of the 90% of the nodules classified as 'indeterminate' on SCT in our study were inadvertently categorised into regenerative or HCC groups on histopathology but could have actually represented dysplastic lesions.
Overall, CT showed a sensitivity and specificity for the diagnosis of malignant nodules of 95% and 77% and for regenerative nodules of 98% and 96%, respectively, to be compared with a previously reported overall CT diagnostic accuracy (approximately 70.5%) for characterisation of cirrhotic nodules [42].
Subsequently, MD parameters were applied to two nodule groups: malignant (comprising of HCC and potentially malignant/indeterminate lesions) and benign (regenerative) nodules, which demonstrated that all parameters could adequately discriminate the two categories with higher diagnostic accuracy of arterial phase parameters. Thus, SCT seems to yield the potential to be an optimum screening modality for dysplastic nodules and to distinguish them from regenerative nodules (Fig. 6). All nodules with ICD nodule HAP values > 11.5 mg/mL and Fig. 6 Spectral computed tomography-based material density (MD) map (iodine only image) and scatter plots with histopathological confirmation of an indeterminate nodule. a Axial MD map on HAP depicting indeterminate nodule with ROI (blue circle) placed at its centre (blue arrow), ROI (red circle) placed on the background liver parenchyma (orange arrow) and ROI (green circle) within the aorta at the same level (green arrow). b Scatter plot demonstrating ICD nodule HAP for nodule (blue arrow) more than the background liver ICD liver HAP (orange arrow) and less than ICD aorta (green arrow) as reference. c Histopathology confirmation of nodule, showing features of high-grade dysplasia. d Axial MD map acquired on PVP showing ROI (red circle) placed in the centre of the nodule (blue arrow), ROI (blue circle) on the liver parenchyma (orange arrow), and aorta (green circle) at the same level (green arrow). e Scatter plot demonstrating ICD nodule PVP (blue arrow) persistently higher than ICD liver PVP (orange arrow). f Immunohistochemistry slide showing CD34-positive foci (black bold arrow) in the nodule depicting dysplasia atypical enhancement pattern could be monitored with a high index of suspicion by the radiologist in tandem with the treating physician for stricter surveillance, till the end point of either the nodule exhibiting classical imaging appearance of HCC or values > 15.5 mg/mL which has been ascertained as a valid cutoff for malignancy with reference to pathology. Although beyond the scope of this study, long-term studies would be required to evaluate and sequentially follow-up dysplastic nodules on SCT, regardless of pathology correlation, to look for metamorphosis into HCC and to determine MD values based on natural disease progression. In addition, 'numerical cutoff values' on SCT would simplify the algorithm for diagnosis of nodules in cirrhosis by facilitating 'operator independent', machine-based learning in the future. We feel that extraction of objective MD values will also open the pathway for demarcation of a variety of other benign (e.g., haemangioma, fibronodular hyperplasia) and non-HCC malignant tumours in the liver.
As our objective was to distinguish nodules in cirrhosis using quantitative MD parameters, we have defined three main MD parameters (ICD nodule HAP, LNR-HAP, and ΔICD) and generated cutoff values for differentiation of malignant or potentially malignant and benign nodules in cirrhosis. These are likely to help radiologists gain confidence while reporting 'lesions in question' by circumventing, operator-dependency, scanner limitations, technical parameters, contrast-injection protocols and hemodynamic changes etc. Spectral CT appears as an interesting functional tool for characterisation of nodules in the cirrhotic liver and likely to open a cemented pathway to radiomics in the near future. The incorporation of MD parameters as quantitative and objective criteria for the adequate distinction of these nodules would help in triaging patients for follow-up, further evaluation or intervention. Early therapeutic measures for potentially malignant lesions will open a new window of opportunities in overall disease survival and prognosis.
There were limitations of our study: lack of an adequate number of dysplastic lesions on histopathology for reference; exclusion of other incidental lesions such as haemangiomas; lipomas, complex cysts; fibronodular hyperplasia; or hepatic adenoma from our study cohort. We also excluded non-HCC tumours and most of the atypical HCC or variants such as combined HCCcholangiocarcinoma, during selection of the patient population. We did not use LI-RADS for initial nodule characterisation in addition to the EASL criteria to designate nodules into more than three categories. In the future, studies are required to assess other hepatic focal lesions such as cysts, mimics of HCC and non-HCC malignancies with quantitative MD parameters for a comprehensive approach. LI-RADS algorithm may also be used in conjunction with MD values for each nodule category in liver cirrhosis.
In conclusion, spectral CT-generated MD parameters, namely ICD nodule HAP, LNR-HAP, and ΔICD, provided a viable diagnostic tool for differentiating malignant or potentially malignant from benign nodules in cirrhotic patients. Such parameters may represent an add-on diagnostic tool for hepatic nodule characterisation.