Introduction
Hepatocellular carcinoma (HCC) is the most common primary hepatic malignant tumor and the third leading cause of cancer-related death worldwide, with annual mortality rates of 14.3 per 100,000 men and 5.1 per 100,000 women.1,2 Hepatic hemangioma (HH) and focal nodular hyperplasia (FNH) are the most and the second most common benign hepatic lesions.3–5 Since the three lesions may share some characteristics upon imaging examination, the differentiation of HCC from HH and FNH is very critical, as their clinical courses, prognosis, and treatment are markedly different.
About 80–90% of HCCs occur as a complication of chronic liver disease, secondary to viral hepatitis B- or C-induced cirrhosis or alcoholic cirrhosis; therefore, patient clinical history may be helpful for the differential diagnosis of HCC from HH and FNH.6–8 Nevertheless, the remaining 10–20% of HCCs can be found in a low-risk population or in patients without alcohol abuse. Besides, typical HCC, HH, and FNH can be diagnosed with confidence using ultrasound, contrast-enhanced computed tomography (CT), or magnetic resonance imaging (commonly known as MRI),9–21 but in some instances, HCC may display atypical imaging presentations if the tumor is well-differentiated,22,23 small (≤2 cm),24 with fatty metamorphosis,25,26 or with abundant interstitial fibrosis.10,27,28 In addition, HH may show slow enhancement or homogeneous enhancement during the arterial phase (AP),29–32 while atypical FNH may show non-enhancement of the central scar, less intense enhancement of the tumor, and pseudo-capsular enhancement in delayed images.33 In the presence of atypical imaging appearances, it may not be easy to distinguish HCC from HH and FNH.
Many studies have shown superior performance of MRI over conventional CT in the diagnosis of HCC;34,35 however, it may be contraindicated in some patients or may result in insufficient image quality in some patients with ascites or patients incapable of holding their breath. With the introduction of dual-energy CT (commonly referred to as DECT) spectral imaging in the field of liver imaging, several studies have shown the benefit of DECT in the improved evaluation of microvascular invasion in HCC,36 in the visualization and quantification of HCC,37 and in the early detection of HCC and hypervascular liver tumors38–40 DECT improved soft tissue contrast by generating different monochromatic images,41,42 which also showed that material decomposed images could provide increased contrast in the visualization of the AP hyperenhancement and washout in HCC compared to both monoenergetic 65 keV images and MRI.37 Until now, only a few studies have reported spectral CT being used in differentiating small HH,43,44 FNH45 or angiomyolipoma46 from HCC with only a few parameters, such as the contrast-to-noise ratio, normalized iodine concentrations (referred to as NIC), and lesion-to-normal parenchyma iodine concentration ratio. Nevertheless, there is almost no literature about systematic and comprehensive comparisons using CT attenuation values derived from a set of monochromatic images (40–140 keV) and other quantitative assessments, including iodine density (referred to as ID), water density (referred to as WD), and the slopes of the spectral curve between HCC, HH, and FNH.
Therefore, this study aimed to describe CT attenuation values derived from a set of monochromatic images and material density-related quantitative assessments for HCC, HH, and FNH, and to evaluate the value of CT spectral imaging in distinguishing HCC from HH and FNH.
Methods
Patients
This retrospective study included patients with known or suspected liver tumors, who underwent dynamic enhancement CT scanning in gemstone spectral imaging (GSI) mode on a Discovery CT750 HD scanner (GE Healthcare, Chicago, IL, USA) between February 2012 and January 2018. The exclusion criteria were: 1) no HCC, HH, or FNH; 2) no histological confirmation; 3) with prior trans-arterial chemoembolization or radiofrequency ablation; or 4) recurrent HCC after liver resection or transplantation. This study was approved by the ethics committee of Ruijin Hospital Affiliated to Shanghai Jiao Tong University School of Medicine. Individual consent was waived because of the retrospective nature of the study.
Diagnostic procedures
All HCCs were confirmed pathologically after surgical resection. The HCCs were graded according to the Edmondson-Steiner classification.47 The diagnosis of HH was established based on histological specimens obtained at partial hepatectomy or typical multiple-phase CT findings, including peripheral nodular enhancement similar to the enhancement of blood vessels at the AP and centripetal fill-in enhancement at the portal venous phase (PP). FNH was proven pathologically after surgical resection because of diagnostic uncertainty after MRI or initial misdiagnosis as HCC.
CT examinations
All patients underwent triple-phase CT within a maximum of 10 days of surgery, using the high-definition Discovery CT750 HD scanner. The detailed scan parameters are provided in the online supplementary materials. Three types of images were reconstructed from the single spectral CT acquisition for analysis, namely conventional polychromatic images obtained at 140 kVp, iodine- and water-based material decomposition (referred to as MD) images, and a set of monochromatic images obtained with energy levels ranging from 40 to 140 keV.
Quantitative analyses
All measurements were performed on an advanced workstation (AW 4.4; GE Healthcare) using the GSI viewer software, by a single radiologist experienced in abdominal radiology and blind to the results of all patients. The 70 keV monochromatic images and iodine-based MD images were reviewed first. Regions of interest (referred to as ROIs) were placed in the lesions, normal liver parenchyma, and aorta on the default 70-keV monochromatic images (Fig. 1). The GSI viewer software automatically calculated the mean CT attenuation value and standard deviation at different energy levels (40–140 keV, at 10-keV intervals) from the set of monochromatic image and ID and WD values from the iodine- and water-based (IDM and WDM, respectively) images during AP and PP.48 Four parameters were obtained from the measurements of CT values and iodine concentration. The NIC during AP was calculated as ratio of the iodine concentrations in AP in the lesions and aorta. Iodine concentrations in the lesions in AP were normalized to those of the aorta in order to minimize variations among patients. The iodine uptake ratio (IUR) was calculated as the ratio of the mean iodine concentrations in the lesions and in the non-tumor hepatic parenchyma surrounding the lesion. The iodine concentration ratio (ICR) in liver lesions between AP and PP was calculated as the difference of the iodine concentrations in the lesions during AP and PP. The slopes of the spectral curve in AP and PP were calculated as (CT40keV−CT90keV)/50, where CT40keV and CT90keV were the mean CT attenuation values in the lesions at 40 keV and 90 keV of the spectral curves, respectively.
Qualitative analysis
Two radiologists (WXL and XTZ, with 15 and 14 years of experience in abdominal CT imaging, respectively) qualitatively reviewed the 70 keV monochromatic images by consensus with the GSI viewer at the workstation. Neither observer was aware of clinical, surgical, and pathologic findings. The observers documented the following enhancement and morphologic features: number; maximal diameter on transverse images; necrosis or cyst; scar; and, enhancement pattern and degree. The enhancement patterns and degrees were evaluated for any enhancing portion of the lesion relative to the aorta and the adjacent liver parenchyma during AP and relative to the adjacent liver parenchyma during PP. The enhancement patterns were described as globular or nodular, diffuse homogeneous, or heterogeneous. The enhancement degree of the lesion was classified as hyper-, iso-, or hypo-enhancement compared with surrounding liver parenchyma. The changes in enhancement degree between AP and PP were characterized as expansion, washout, or none. The expansion was defined as a hyperenhancement area in the lesion during both AP and PP. Washout was defined as a change from hyper- or iso-enhancement area in the lesion during AP to a hypo-enhancement area in the lesion compared with surrounding liver parenchyma during PP, while none was defined as a hyper- or iso-enhancement area in the lesion during AP to an iso-enhancement area in the lesion compared with adjacent liver parenchyma during PP.
Finally, the observers, in consensus, characterized each lesion type as HCC, HH, and FNH based on imaging features (online supplementary materials). Differences among the observers were resolved by means of consensus discussion. The definition of the sensitivity and specificity for differential diagnosis of HCC, HH and FNH are demonstrated in the online supplementary materials.
Statistical analysis
The data were analyzed using SPSS 13.0 (SPSS Inc., Chicago, IL, USA). Continuous variables were presented as median (interquartile range) after confirming their non-normal distribution using the Kolmogorov-Smirnov test. The estimated parameters were analyzed among HCC, HH, and FNH groups using the Kruskal-Wallis test, with the Wilcoxon rank-sum post hoc test. Receiver operating characteristic (ROC) curves were used to help establish the threshold values of the mean CT values at different energy levels (40–140 keV), IDM, WDM, IUR and slope in AP, and PP, and ICR required for significant differentiation of HCC, HH, and FNH. The diagnostic capability was determined by calculating the area under the ROC curve (AUC). The best sensitivity and specificity were determined using the optimal thresholds based on the Youden’s index. A two-sided p value of <0.05 was considered statistically significant. Qualitative CT imaging features were compared among HCC, HH, and FNH by Fisher’s exact tests.
Results
Patients
From 476 potentially eligible patients, 425 were excluded based on the following exclusion criteria: 1) no HCC, HH, or FNH; 2) no histological confirmation; 3) with prior transarterial chemoembolization or radiofrequency ablation; or 4) recurrent HCC after liver resection or transplantation. Therefore, 51 patients (30 males and 21 females) were included, with 31 HCCs in 31 patients, 17 HHs in 13 patients, and 7 FNHs in 7 patients (Table 1). According to the Edmondson-Steiner classification, two HCCs were grade I, fourteen were grade II, thirteen were grade III, and two were grade IV. All patients who were diagnosed as HH with multiple-phase CT findings were followed for at least 3 years. No patient was lost to follow-up.
Table 1Characteristics of the patients
Variable | HCC, n=31 | HH, n=13 | FNH, n=7 |
---|
Number of lesions | 31 | 17 | 7 |
Age in years, median (IQR) | 57.0 (16.0) | 48.0 (18.0) | 24.0 (11.0) |
Male, n (%) | 26 (83.4) | 2 (15.4) | 2 (28.6) |
Background liver status, n (%) | | | |
Normal liver | 2 (6.5) | 13 (100) | 7 (100) |
Cirrhosis | 29 (93.5) | 0 | 0 |
Cause, n (%) | | | |
HBV infection | 28 (90.3) | 0 | 0 |
HBV and alcoholic cirrhosis | 1 (9.7) | 0 | 0 |
Tumor size in cm, median (IQR) | 3.00 (3.00) | 5.30 (3.35) | 5.00 (2.00) |
Quantitative analysis
CT values
Regarding the CT values of HCC, HH, and FNH, there was a trend towards a decrease in mean CT values of HH, HCC, and FNH as energy level increased (40–140 keV) in both AP and PP (Table 2 and Fig. 2). HH had the lowest mean CT values, while FNH had the highest mean CT values at different energy levels (40–140 keV) in both AP and PP (Table 2 and Fig. 2). Moreover, there were significant differences in CT values at energy levels from 40 to 140 keV during AP and PP (Table 2) for HCC vs. FNH, and HH vs. FNH. There were significant differences between HCC and HH at energy levels from 40 to 140 keV in AP and only at energy levels from 40 to 100 keV in PP (Table 2).
Table 2Mean CT attenuation values of HCC, HH, and FNH at energy levels ranging 40–140 keV (at 10-keV intervals) during AP and PP
Energy in keV | HH, n=17 | HCC, n=31 | FNH, n=7 | p
|
---|
HCC vs. HH | HCC vs. FNH | FNH vs. HH |
---|
AP | | | | | | |
40 | 102.77 (57.02) | 210.02 (125.86) | 318.74 (165.34) | <0.001* | 0.004* | <0.001* |
50 | 81.67 (36.37) | 156.48 (83.43) | 224.71 (113.64) | <0.001* | 0.003* | <0.001* |
60 | 66.45 (23.41) | 115.87 (57.17) | 166.84 (71.39) | <0.001* | 0.006* | <0.001* |
70 | 59.31 (17.24) | 95.09 (41.81) | 130.42 (42.74) | <0.001* | 0.001* | <0.001* |
80 | 54.65 (15.29) | 81.35 (29.19) | 113.65 (26.49) | <0.001* | <0.001* | <0.001* |
90 | 48.65 (15.52) | 71.15 (23.73) | 103.64 (16.13) | <0.001* | <0.001* | <0.001* |
100 | 47.45 (16.05) | 65.59 (22.63) | 90.07 (15.13) | <0.001* | 0.001* | <0.001* |
110 | 46.99 (15.41) | 61.76 (20.84) | 78.06 (17.20) | 0.002* | 0.001* | <0.001* |
120 | 45.56 (15.06) | 58.96 (20.06) | 69.62 (18.75) | 0.004* | 0.001* | <0.001* |
130 | 44.08 (14.76) | 56.62 (17.53) | 65.52 (21.27) | 0.008* | 0.002* | <0.001* |
140 | 42.89 (14.21) | 53.91 (16.29) | 62.56 (23.54) | 0.018* | 0.004* | <0.001* |
PP | | | | | | |
40 | 179.57 (89.33) | 246.44 (78.64) | 338.48 (63.72) | 0.001* | 0.004* | 0.001* |
50 | 131.95 (58.34) | 176.12 (50.17) | 238.85 (38.62) | 0.001* | 0.002* | 0.001* |
60 | 102.98 (40.96) | 132.85 (36.90) | 177.56 (32.60) | 0.002* | 0.001* | 0.001* |
70 | 84.44 (28.58) | 105.22 (25.08) | 142.38 (28.65) | 0.003* | 0.001* | 0.001* |
80 | 72.55 (22.82) | 87.80 (19.26) | 114.83 (26.14) | 0.004* | <0.001* | <0.001* |
90 | 64.49 (19.53) | 74.59 (16.54) | 98.60 (24.60) | 0.010* | <0.001* | <0.001* |
100 | 59.04 (18.46) | 66.74 (16.56) | 87.38 (23.48) | 0.036* | <0.001* | <0.001* |
110 | 54.73 (17.71) | 60.70 (16.34) | 79.44 (22.23) | 0.065 | <0.001* | <0.001* |
120 | 50.80 (16.64) | 56.48 (15.14) | 73.88 (21.25) | 0.113 | 0.001* | <0.001* |
130 | 47.87 (15.76) | 53.57 (13.83) | 69.62 (21.17) | 0.185 | 0.001* | <0.001* |
140 | 47.05 (15.01) | 50.92 (13.13) | 66.25 (21.36) | 0.200 | 0.002* | 0.001* |
ROC curve analysis revealed that the mean CT values from 40–140 keV in both AP and PP, especially at 40–120 keV in AP (all AUC=1) and 120 keV in PP (AUC=0.992) had the best performance in differentiating HH from FNH. The mean CT values at 80–90 keV in AP (both AUC=0.926) and 100 keV in PP (AUC=0.949) had better performance in differentiating HCC from FNH. Meanwhile, the mean CT values at 40–50 keV in AP (AUC=0.896) and 40–50 keV in PP (AUC=0.780) had high sensitivity and specificity in differentiating HCC from HH. See Supplementary Tables S1–S3.
Standard deviation of the mean CT values
For the standard deviation of the mean CT values of HCC, HH, and FNH in AP and PP (Table 3), there were significant differences at 40 to 100 keV during PP between HCC and HH, and from 40 to 140 keV during PP between HH and FNH. In addition, sensitivity (88.2%) and specificity (100%) showed that the standard deviation of mean CT values at 40 keV in PP (AUC=0.882) had the best performance in differentiating HH from FNH. The standard deviation of mean CT values from 40–140 keV in PP showed low sensitivity and specificity in differentiating between HCC and FNH. The standard deviation of mean CT values at 40–100 keV in PP, especially at 50 keV (AUC=0.846), had the best sensitivity and specificity in differentiating between HCC and HH. See Supplementary Tables S4–S6.
Table 3Standard deviations of mean CT attenuation values (the heterogeneity of lesions in mean CT values) of HCC, HH, and FNH at energy levels ranging 40–140 keV (at 10-keV intervals) during AP and PP
Energy in keV | HH, n=17 | HCC, n=31 | FNH, n=7 | p
|
---|
HCC vs. HH | HCC vs. FNH | FNH vs. HH |
---|
AP | | | | | | |
40 | 74.15 (54.18) | 62.04 (20.15) | 58.32 (47.92) | 0.407 | 0.985 | 0.589 |
50 | 48.04 (35.43) | 42.05 (14.65) | 41.27 (36.32) | 0.438 | 0.778 | 0.727 |
60 | 33.63 (23.01) | 31.79 (12.21) | 34.00 (24.50) | 0.659 | 0.665 | 0.924 |
70 | 23.63 (15.87) | 21.77 (7.81) | 23.16 (17.35) | 0.643 | 0.624 | 0.775 |
80 | 21.84 (7.10) | 20.42 (5.83) | 19.89 (10.86) | 0.316 | 0.585 | 0.727 |
90 | 23.60 (4.95) | 19.87 (8.73) | 22.82 (7.54) | 0.152 | 0.721 | 0.727 |
100 | 24.24 (6.80) | 19.71 (12.44) | 22.88 (5.92) | 0.337 | 0.836 | 0.193 |
110 | 23.00 (7.59) | 20.26 (14.06) | 21.16 (4.60) | 0.407 | 0.985 | 0.216 |
120 | 23.23 (8.71) | 19.52 (12.42) | 19.82 (3.55) | 0.525 | 0.778 | 0.172 |
130 | 23.45 (9.56) | 19.64 (12.36) | 18.73 (2.92) | 0.553 | 0.679 | 0.266 |
140 | 23.67 (10.21) | 19.44 (13.13) | 17.92 (2.42) | 0.511 | 0.638 | 0.357 |
PP | | | | | | |
40 | 125.30 (73.50) | 52.41 (30.98) | 54.04 (18.74) | <0.001* | 0.778 | 0.001* |
50 | 81.20 (46.81) | 35.60 (16.82) | 39.27 (11.48) | <0.001* | 0.440 | 0.001* |
60 | 54.92 (31.65) | 25.12 (13.28) | 31.96 (6.28) | <0.001* | 0.463 | 0.001* |
70 | 39.25 (21.45) | 17.61 (8.43) | 21.41 (4.70) | <0.001* | 0.283 | 0.002* |
80 | 32.81 (13.96) | 18.22 (5.73) | 18.30 (3.85) | <0.001* | 0.865 | 0.001* |
90 | 28.89 (8.62) | 21.07 (10.46) | 20.15 (5.14) | <0.001* | 0.463 | 0.001* |
100 | 27.20 (5.05) | 21.02 (10.67) | 19.20 (6.34) | 0.011* | 0.275 | 0.001* |
110 | 24.88 (6.76) | 20.76 (11.82) | 17.73 (6.82) | 0.054 | 0.251 | 0.003* |
120 | 24.86 (8.45) | 20.67 (13.16) | 16.73 (7.17) | 0.200 | 0.236 | 0.006* |
130 | 24.72 (10.42) | 20.86 (14.19) | 15.98 (7.39) | 0.258 | 0.236 | 0.010* |
140 | 24.31 (11.75) | 20.01 (15.08) | 15.44 (7.55) | 0.337 | 0.221 | 0.010* |
Quantitative assessments
For the spectral CT imaging-specific quantitative assessments of HCC, HH, and FNH, there were significant differences in IUR, slope in AP and PP, NIC in AP (which tended to increase from HH, HCC to FNH) between every two groups of HCC, HH, and FNH (Table 4). IDM in both AP and PP, WDM in PP, and ICR revealed significant differences between HH and FNH, as well as IDM in AP and PP, and ICR between HH and HCC, WDM in PP between HCC and FNH also differed significantly (Table 4). ROC curve analysis demonstrated that ICR and slope in AP had the best diagnostic performance (AUC=0.992) in differentiating HH from FNH. While IUR in AP had better performance (AUC=0.903) in differentiating HCC from FNH. For distinguishing HCC from HH, IDM and slope in AP, ICR had good performance (AUC=0.890). See Supplementary Tables S7–S9.
Table 4Quantitative assessment of HCC, HH and FNH with CT spectral imaging
| HH, n=17 | HCC, n=31 | FNH, n=7 | p
|
---|
HCC vs. HH | HCC vs. FNH | FNH vs. HH |
---|
IDM in the AP (water mg/mL) | 0.75 (0.70) | 2.04 (1.51) | 3.34 (2.56) | <0.001* | 0.080 | 0.006* |
IDM in the PP (water mg/mL) | 1.73 (1.00) | 2.56 (1.17) | 3.53 (1.09) | 0.002* | 0.052 | 0.017* |
WDM in the AP (iodine mg/mL) | 1,037.77 (11.12) | 1,038.99 (12.22) | 1,046.02 (20.98) | 0.872 | 0.087 | 0.105 |
WDM in the PP (iodine mg/mL) | 1,038.22 (16.80) | 1,035.67 (12.53) | 1,047.36 (22.31) | 0.643 | 0.010* | 0.039* |
ICR | 0.45 (0.18) | 0.81 (0.26) | 0.95 (0.10) | <0.001* | 0.118 | <0.001* |
IUR in the AP | 1.22 (0.61) | 2.70 (1.75) | 6.94 (6.39) | <0.001* | 0.001* | <0.001* |
IUR in the PP | 0.57 (0.30) | 0.93 (0.32) | 1.18 (0.29) | 0.002* | 0.009* | 0.002* |
NIC in the AP | 0.06 (0.06) | 0.17 (0.13) | 0.29 (0.08) | <0.001* | 0.002* | 0.002* |
Slope in the AP | 1.01 (0.98) | 2.78 (2.06) | 4.54 (2.45) | <0.001* | 0.008* | <0.001* |
Slope in the PP | 2.34 (1.36) | 3.48 (1.58) | 4.80 (1.32) | 0.007* | 0.004* | 0.001* |
Qualitative analysis
The CT features of the HH, HCC, and FNH groups were analyzed and listed in Table 5. Feeding vessels were found in 15 (48.4%) of the 31 HCCs, 6 (85.7%) of the 7 FNHs, and none was found in HHs (both p<0.001 for HCC vs. HH, FNH vs. HH, respectively). Thirteen (76.5%) of seventeen HHs showed globular or nodular enhancement during AP, whereas neither HCCs nor FNHs showed it (both p<0.001 for HH vs. HCC and for HH vs. FNH). Fourteen (82.4%) of seventeen HHs demonstrated expansion change of enhancement between AP and PP (p<0.001 for HH vs. HCC and p=0.001 HCC vs. FNH, respectively), whereas twenty-two (71%) of the thirty-one lesions of HCC demonstrated washout change of enhancement between AP and PP (p<0.001 for HH vs. HCC and p=0.001 for HCC vs. FNH, respectively).
Table 5Qualitative CT assessment of lesions
Feature | HH, n=17 | HCC, n=31 | FNH, n=7 | p
|
---|
HCC vs. HH | HCC vs. FNH | FNH vs. HH |
---|
Necrosis or cyst | 2 (11.8) | 7 (22.6) | 0 (0) | 0.460 | 0.309 | 1 |
Central scars | 2 (11.8) | 0 (0) | 4 (57.1) | 0.121 | 0.001* | 0.174 |
Presence of feeding vessels | 0 (0) | 15 (48.4) | 6 (85.7) | <0.001* | 0.104 | <0.001* |
Pseudocapsule | 0 (0) | 8 (25.8) | 3 (42.9) | 0.038 | 0.390 | 0.017* |
Attenuation of enhancing area in AP | | | | | | |
Similar to aorta | 7 (41.2) | 0 (0) | 0 (0) | <0.001* | – | 0.065 |
Less than aorta, greater than adjacent liver parenchyma | 10 (58.8) | 31 (100) | 7 (100) | <0.001* | – | 0.065 |
Enhancement pattern in AP | | | | | | |
Globular or nodular | 13 (76.5) | 0 (0) | 0(0) | <0.001* | – | <0.001* |
Diffuse homogeneous | 4 (23.5) | 10 (32.3) | 5 (71.4) | 0.741 | 0.089 | 0.061 |
Diffuse heterogeneous | 0 (0) | 21 (67.7) | 2 (28.6) | <0.001* | 0.089 | 0.076 |
Enhancement change between AP and PP | | | | | | |
Expansion | 14 (82.4) | 0 (0) | 4 (57.1) | <0.001* | 0.001* | 0.307 |
Washout | 0 (0) | 22 (71) | 0 (0) | <0.001* | 0.001* | – |
None | 3 (17.6) | 9 (29) | 4 (42.9) | 0.497 | 0.203 | 0.134 |
In the conventional qualitative analysis of imaging features with combined AP and PP, we achieved sensitivity of 83.9% (26 of 31 HCCs) and specificity of 82.4% (14 of 17 HHs), respectively, for differentiating between HCC and HH, which was lower than quantitative image analysis with CT spectral imaging, which had sensitivity of 87.1% and specificity of 88.2%. Meanwhile, for differentiating between FNH and HH, we achieved sensitivity of 76.5% (13 of 17 HHs) vs. 100%, specificity 71.4% (5 of 7 FNHs) vs. 100%. In addition, we achieved sensitivity of 80.6% (25 of 31 HCCs) vs. 100%, and specificity 71.4% (5 of 7 FNHs) vs. 90.3%, respectively.
Discussion
The present study suggests that spectral CT can be helpful for differentiating HCC from HH and FNH. HH had the lowest mean CT values, while FNH had the highest mean CT values at different energy levels. The different slopes clearly indicated that the CT spectral imaging could distinguish HCC, HH, and FNH.
Treatment options and the prognosis of HCC, HH, and FNH are markedly different. Previous studies discussed the role of CT in the diagnosis of HCC, HH, and FNH,9,11,19,20,30–33 but they were primarily focusing on analyzing the imaging features of the lesions qualitatively. Some studies reported quantitative evaluations of spectral CT in differentiating small HH,43 FNH,45 or angiomyolipoma46 from HCC with only a few parameters. This study systematically and comprehensively compared CT values derived from a set of monochromatic images (from 40 to 140 keV at 10-keV intervals), and quantitative assessments including ID, WD, and the slopes of the spectral curve for differentiating HCC, HH, and FNH. The study expands the results of previous studies43,45,46 that only examined spectral CT features of HCC vs. one lesion type and with limited energy levels. This is supported by Wang et al.,48 who showed that spectral CT features could differ between malignant and benign liver lesions.
The present study shows that the mean CT values and the standard deviation of mean CT values measured on monochromatic images at low energy levels (40–90 keV), especially in AP, have better contrast resolution compared to monochromatic images at high energy levels (100–140 keV). This means that the CT values and their standard deviation (which represent the heterogeneity of lesions in CT values) measured on monochromatic images at certain energy levels could be helpful for the differential diagnosis of different hypervascular hepatic tumors. This is partly similar to the studies reported by Lv et al.40 and Yu et al.,45 who found that monochromatic images at energy levels of 40–70 keV could improve the differential diagnosis of small HCC compared to conventional polychromatic images. ROC curve analysis in the study also confirmed that quantitative analysis of CT spectral data had higher sensitivity and specificity with those of conventional qualitative CT image analysis during combined phases for differentiating HCC, FNH and HH.
According to the present study results, all values of IUR in the AP for HCC, FNH, and HH were >1, which means that all three groups showed hyperenhancement appearance during AP. The value of IUR in the PP of HCC was <1 during PP, which was in line with the characteristics of “washout” feature of typical HCC during PP. While the value of IUR of FNH >1 during PP conformed to persistent enhancement characteristics of most FNH during PP. On the other hand, the value of IUR <1 in PP for HH, was contrary to the typically persistent enhancement feature. This could be due to the measurement of IDM in PP, including the part of HH without iodine filling during AP or PP. The ROI encompassed as much of the lesions as possible in order to measure the heterogeneity of lesion, and this caused the IUR in PP (IDM in PP of the lesion divided by the IDM in the PP of the liver) to be >1.
The present study did have some limitations. First, this was a retrospective study, with all the biases inherent to that study design. Second, this was a preliminary study with a small number of patients and needs to be verified by additional studies performed with a larger number of patients. Third, the HCCs in the present study were not classified by histopathological grade, because the numbers of patients in grades I and IV were too small. Fourth, because only one reader examined the images, intra- or inter-observer variability data were lacking. Fifth, and most importantly, there was a lack of correlation with conventional MDCT morphologic findings and typical features. Finally, quantitative analysis is time-consuming, and many of the described parameters could not be quantitated on picture archiving and communication system (PACS) and would need a dedicated AW workstation. Additional studies are necessary to address these issues.
Conclusions
In conclusion, spectral CT provides a set of monochromatic images, iodine-based MD images, and the quantitative parameters based on iodine concentration analysis that may help to increase the accuracy of the differentiation of HCC, HH, and FNH.
Supporting information
Supplementary Methods
Detailed scan parameters of CT examination.
(DOCX)
Supplementary Table S1
ROC values of differentiating HH from FNH with CT attenuation value at 40-140 keV (at 10-keV intervals).
(DOCX)
Supplementary Table S2
ROC analysis of differentiating HCC from FNH with CT attenuation value at 40-140 keV (at 10-keV intervals).
(DOCX)
Supplementary Table S3
ROC values of differentiating HCC from HH with CT values at 40-140keV (at 10-keV intervals).
(DOCX)
Supplementary Table S4
ROC values of differentiating HH from FNH with the standard deviation of CT values at 40-140 keV (at 10-keV intervals) in PP.
(DOCX)
Supplementary Table S5
ROC values of differentiating HCC from FNH with the standard deviation of CT values at 40-140 keV (at 10-keV intervals) in PP.
(DOCX)
Supplementary Table S6
ROC values of differentiating HCC from HH with standard deviation of CT values at 40-140keV (at 10-keV intervals) in PP.
(DOCX)
Supplementary Table S7
ROC analysis of differentiating HH from FNH using spectral CT imaging-specific parameters.
(DOCX)
Supplementary Table S8
ROC analysis of differentiating HCC from FNH using spectral CT imaging-specific parameters.
(DOCX)
Supplementary Table S9
ROC analysis of differentiating HCC from HH using spectral CT imaging-specific parameters.
(DOCX)
Abbreviations
- AP:
arterial phase
- AUC:
the area under the ROC curve
- CNR:
contrast-to-noise ratio
- CT:
computed tomography
- DECT:
dual-energy computed tomography
- FNH:
focal nodular hyperplasia
- GSI:
gemstone spectral imaging
- HCC:
hepatocellular carcinoma
- HH:
hepatic hemangioma
- ICR:
iodine concentration ratio
- ID:
iodine density
- IUR:
iodine uptake ratio
- LNR:
lesion-to-normal parenchyma ratio
- MD:
material decomposition
- MRI:
magnetic resonance imaging
- NIC:
normalized iodine concentrations
- PACS:
picture archiving and communication system
- PP:
portal venous phase
- ROC:
receiver operating characteristic
- ROI:
Regions of interest
- WD:
water density
Declarations
Acknowledgement
The authors are grateful to Dr. Jianying Li at GE Healthcare in revising the manuscript.
Data sharing statement
No additional data are available.
Funding
This work was supported by the National Natural Science Foundation of China (Grant number 81401406) and the Innovative Research Team of High-Level Local Universities in Shanghai.
Conflict of interest
The authors have no conflict of interests related to this publication.
Authors’ contributions
Carried out the studies, participated in collecting data and drafted the manuscript (WL), carried out the studies and participated in collecting data (RL, XZ), participated in collecting data (YY, JZ), performed the statistical analysis and participated in its design (XL, WC), participated in acquisition and analysis of data (KC), participated in interpretation of data (FY). Manuscript writing and final approval of manuscript (all authors).