Value of intravoxel incoherent motion and diffusion kurtosis imaging to differentiate hepatocellular carcinoma and intrahepatic cholangiocarcinoma

Abstract

BACKGROUND

The differential diagnosis between hepatocellular carcinoma (HCC) and intrahepatic cholangiocarcinoma (ICC) is crucial. The individual differences of patients increase the complexity of diagnosis. Currently, imaging diagnosis mainly relies on conventional computed tomography and magnetic resonance imaging (MRI), but few studies have investigated MRI functional imaging. This study combined MRI functional imaging including intravoxel incoherent motion (IVIM) and diffusion kurtosis imaging (DKI), facilitating differential diagnosis.

AIM

To explore the differential diagnostic value of IVIM imaging and DKI in differentiating between HCC and ICC.

METHODS

A total of 58 patients who underwent multi-b-value diffusion weighted imaging (DWI) on a 3.0 T magnetic MRI scanner were enrolled in this study. Standard apparent diffusion coefficient (SADC), IVIM quantitative parameters, including pure diffusion coefficient (D), pseudo diffusion coefficient (Dstar), and perfusion fraction (f), as well as the DKI quantitative parameters mean diffusion coefficient (MD) and mean kurtosis coefficient (MK) were computed by multi-b DWI images. The χ² test was used for classified data, and a one-way analysis of variance was performed for counted data. P < 0.05 indicated statistical significance. The diagnostic value of parameters in HCC and ICC was analyzed using the receiver operating characteristic (ROC) curve.

RESULTS

The SADC, D, and MD values were significantly lower in the HCC group compared to the ICC group, whereas MK was significantly higher in the HCC group than in the ICC group (P < 0.05). No significant difference in Dstar and f was observed between the HCC group and the ICC group (P > 0.05). The optimal cutoff levels of the total values of SADC, D, MK, MD and all associated parameters were 1.25 × 10^-3 mm²/second, 1.32 × 10^-3 mm²/second, 650.2 × 10^-3 mm²/second, 1.41 × 10^-3 mm²/second and 0.46 × 10^-3 mm²/second, respectively. The sensitivity of diagnosis was 95%, 80%, 90%, 100%, and 70%, respectively, the specificity of diagnosis was 67.39%, 69.57%, 67.39%, 43.48%, and 93.48%, respectively, and the area under the ROC curve was 0.874, 0.793, 0.733, 0.757, and 0.895, respectively.

CONCLUSION

SADC, D, MK, and MD could be used to distinguish HCC from ICC, with the diagnostic value reaching a maximum after establishing a joint model.

Key Words: Hepatocellular carcinoma; Magnetic resonance imaging; Intrahepatic cholangiocarcinoma; Intravoxel incoherent motion; Diffusion kurtosis imaging

Core Tip: This study employed intravoxel incoherent motion (IVIM) and diffusion kurtosis imaging (DKI) for liver cancer and intrahepatic cholangiocarcinoma. The findings suggest that the parameters standard apparent diffusion coefficient (SADC) and pure diffusion coefficient from IVIM and mean kurtosis coefficient and mean diffusion coefficient from DKI, can differentiate between the two tumor types. Among them SADC demonstrated the highest diagnostic value, and particularly when used in the joint model. The innovation of this lies in the application of the DKI sequence for the first time and analyze its value.

INTRODUCTION

Hepatocellular carcinoma (HCC) and intrahepatic cholangiocarcinoma (ICC) are the most common primary malignant liver tumors. HCC originates from hepatocytes, while ICC originates from malignant tumors in the secondary bile ducts of the liver extending to the epithelium of the peripheral bile ducts. Compared to HCC, ICC is more invasive and presents higher metastatic potential, with a higher mortality[1,2]. Therefore, their treatment methods and prognoses differ significantly, emphasizing the importance of differential diagnosis.

Using traditional imaging diagnostic methods, HCC is characterized by high enhancement in the arterial phase and washout in the portal vein phase and delayed phase, often accompanied by capsule signs. Conversely, ICC is primarily characterized by a slight increase in the solid part of the tumor in the arterial phase and centripetal enhancement in the portal vein phase, accompanied by central necrosis, shrinkage of the adjacent liver capsule, and dilation of the bile duct around the tumor. However, due to heterogeneity, tumors do not always present typical manifestations. According to LI-RADS version 2018[3], the combination of hepatitis B surface antigen positivity, elevated alpha-fetoprotein (AFP), and carbohydrate antigen 19-9 (CA 19-9) levels can differentiate HCC from ICC. However, recent studies have shown that ICC can also be associated with hepatitis B or liver cirrhosis, accompanied by abnormal laboratory indicators related to HCC[4-7], further complicating the differential diagnosis between the two.

Previous studies have attempted to use magnetic resonance imaging (MRI) to differentiate HCC from ICC by combining the enhancement pattern of hepatocyte-specific contrast agents with diffusion-weighted imaging. The traditional single-exponential diffusion weighted imaging (DWI) model has been employed for the diagnosis, differential diagnosis, and to determine treatment response and prognosis of liver tumors[8,9]. Nonetheless, this imaging modality is based on a simple Gaussian diffusion hypothesis and has inherent limitations. In contrast, the emerging non-Gaussian diffusion model provides various biological insights from different parameters. Among these, the intravoxel incoherent motion (IVIM) model is a multi-exponential model that defines the relationship between signal attenuation and increasing b-value. In this model, blood perfusion is distinguished from the actual diffusion of water molecules, thereby deriving multiple parameters. The pure diffusion coefficient (D) reflects the true diffusion of water molecules, the pseudo diffusion coefficient (Dstar) reflects the incoherent motion of microcirculation within voxels, and the perfusion fraction (f) represents the volume ratio of the perfusion effect caused by local microcirculation to the total diffusion effect[10].

Additionally, the non-Gaussian diffusion model, diffusion kurtosis imaging (DKI), quantifies the non-Gaussian behavior of diffusion by analyzing the mean kurtosis coefficient (MK) and mean diffusion coefficient (MD). This model provides new in vivo diffusion characteristics to describe the microstructure of tissues, measuring the extent to which tissue diffusion deviates from the Gaussian model. Currently, IVIM and DKI are applied in the pathological grading of HCC[11,12], evaluation of treatment response[13-15], assessment of microvascular invasion (MVI)[16-18], prediction of independent risk factors affecting liver cancer recurrence, and prediction of survival rates[14,19-22]. However, the application of IVIM in differentiating HCC from ICC is still in its early stages[23-27], and the results have been inconsistent. Furthermore, DKI has not yet been studied for this purpose.

This study aimed to demonstrate the reliability and stability of the non-Gaussian models IVIM and DKI in the differential diagnosis of HCC and ICC by analyzing the parameters derived from these models.

MATERIALS AND METHODS

Study population

This study was approved by the local Institutional Review Board. Informed consent was obtained from all patients for this retrospective review. Between May 2023 and June 2024, 220 consecutive patients were suspected of having chronic liver disease or focal hepatic lesions based on clinical history or previously performed sonography or computed tomography scans in our department. The inclusion criteria were: (1) MRI scanning was performed within two weeks before treatment, including routine scanning, enhanced scanning, and multi-b-value DWI sequences; (2) No treatment history before the examination; (3) The MRI scan was followed by surgical resection or fine needle aspiration biopsy, and the tumor was pathologically confirmed as HCC or ICC; (4) The images had no artifacts; and (5) The diameter of the lesion was more than 1 cm. The diagnosis was based on the World Health Organization Classification of Biliary Malignancies, and one case of mixed HCC was excluded. A total of 65 cases were confirmed by pathology as HCC (39 cases surgically removed, 8 cases confirmed by fine needle biopsy) or ICC (6 cases surgically removed, 12 cases confirmed by fine needle biopsy). Seven cases were excluded due to image motion artifacts. Finally, 58 patients (40 HCC and 18 ICC, comprising 43 men and 15 women) were included in the study. The age range was 49–86 years, with a mean age of 73.4 years. The treatment background and laboratory examinations of all patients were reviewed, including routine blood tests, infectious disease tests, liver function tests, and tumor markers [AFP, abnormal prothrombin, CA 19-9, and carcinoembryonic antigen (CEA)], as well as tumor diameter and Child-Pugh score.

MRI examination

All patients underwent MRI examinations using a 3.0 T MRI scanner (790, United Imaging Healthcare, Shanghai, China) with a 24-channel Superflex body coil. The multi-b-value DWI sequence was acquired using the echo-planar imaging technique. The sequence parameters were: (1) Repetition time (TR)/echo time (TE) = 3224/66 msec; (2) Flip angle (FA) = 90°; (3) Field of view (FOV) = 380 mm × 300 mm; (4) Slice thickness = 6.5 mm; (5) Scanning matrix = 112 × 100; (6) The b-values (average) = 0 (1) mm²/second, 20 (1) mm²/second, 40 (1) mm²/second, 60 (1) mm²/second, 80 (1) mm²/second, 150 (1) mm²/second, 200 (2) mm²/second, 400 (2) mm²/second, 800 (4) mm²/second, 1500 (6) mm²/second, and 2000 (8) mm²/second; (7) Gradient directions = 3 orthogonal directions; and (8) Scan time = 320 seconds. Other routinely acquired MRI sequences included a T2-weighted fast spin-echo sequence (TR/TE = 3764/98 msec; FA = 90°; FOV = 380 mm × 300 mm; slice thickness = 6.5 mm; scanning matrix = 304 × 80; and scan time = 38 seconds), an in-phase and out-phase gradient echo (GRE)-based T1-weighted imaging (T1WI) (TR/TE1/TE2 = 3.96/1.39/2.32 msec; FA = 12°; FOV = 400 mm × 280 mm; slice thickness = 3 mm; scanning matrix = 304 × 202; and scan time = 16 seconds), and contrast-enhanced GRE-based T1WI (TR/TE = 4.78/2.24 msec; FA = 12°; FOV = 400 mm × 280 mm; slice thickness = 3 mm; scanning matrix = 288 × 144; and scan time = 16 seconds). Arterial, portal venous, and delayed phases were obtained at 20–30 seconds, 70–80 seconds, and 180 seconds after a 0.2 mL/kg bolus injection of gadopentetate dimeglumine at a rate of 2 mL/second.

Data post-processing

The original data was transferred to the unresponsive wakefulness syndrome-magnetic resonance (MR) workstation synchronously and post-processed using the Diffusion Analysis software package.

IVIM

The fitting formula for IVIM was as follows: Sb/S0 = (1 − f) × exp × (− b × D) + f × exp × (− b × Dstar) where S0 and Sb represent the signal intensities at b = 0 mm²/second and other b-values, respectively. Dstar and D represent the Dstar and true diffusion coefficient, while f represents the perfusion fraction. Dstar and D are related to perfusion-related diffusion and pure molecular diffusion, respectively. F indicates the fraction of diffusion linked to microcirculation. Moreover, the fitting of IVIM was performed using images with nine b-values (0 mm²/second, 20 mm²/second, 40 mm²/second, 60 mm²/second, 80 mm²/second, 150 mm²/second, 200 mm²/second, 400 mm²/second, and 800 mm²/second). The standard apparent diffusion coefficient (SADC) maps were generated using all nine b-values mentioned above.

DKI

The fitting formula for DKI was as follows: Sb/S0 = −b × MD + 1/6 × MD2 × b2 × MK. Where S0 and Sb are similar to those in IVIM. MD indicates the overall apparent diffusion within the voxel, while MK characterizes the complexity of the organizational microstructure. The fitting of DKI was performed using images with four b-values (0 mm²/second, 800 mm²/second, 1500 mm²/second, and 2000 mm²/second).

All DWI, IVIM, and DKI parameters were measured using the software provided by the supplier. Two abdominal radiologists with 7 years and 10 years of experience independently analyzed the data. The region of interest (ROI) was manually drawn on axial DWI (b = 800 mm²/second) images, referring to conventional T2-weighted imaging images before placing the ROI. While delineating the ROIs, large necrotic areas, bleeding, blood vessels, and bile ducts were avoided, ensuring placement on the largest cross-section of the tumor. Each ROI was placed at least 1 mm away from the edge of the tumor to minimize the partial volume effect. The placement of all ROIs was blinded to clinical data. The corresponding combination of b-values for IVIM and DKI was selected to obtain the parameter values and rainbow diagrams for IVIM and DKI, respectively.

Statistical analysis

All analyses were conducted using Statistical Package for the Social Sciences version 20.0 (SPSS, Chicago, IL, United States) and MedCalc version 15.8 (MedCalc, Mariac, Belgium). All numerical data are expressed as mean ± SD. The clinical characteristics of HCC and ICC patients were analyzed using an independent sample t-test for continuous variables and a χ² test for categorical variables. In addition, the differences in MR parameters between ICC and HCC were compared using an independent sample t-test. A Bland-Altman analysis was performed to assess the consistency of measurements between the two observers. Receiver operating characteristic (ROC) curve analysis was used to calculate the area under the ROC curve (AUC), sensitivity, specificity, 95%CI, and threshold points of the parameters extracted from IVIM and DKI values. In this study, a P < 0.05 was considered significant.

RESULTS

Patient characteristics

Among the 58 patients included in the analysis, the HCC group comprised 35 males and 5 females, with an average age of 59.95 years ± 9.32 years. The ICC group comprised 8 males and 10 females, with an average age of 61.72 years ± 8.17 years. The P values for the differences in age and sex between the two groups were 0.409 and 0.001, respectively. The HCC group comprised 38 cases of Child-Pugh A and 2 cases of Child-Pugh B, whereas the ICC group included 15 cases of Child-Pugh A and 3 cases of Child-Pugh B, showing no statistically significant difference between the two groups (P = 0.143). A total of 66 tumors were detected among the 58 patients. The size of the HCC and ICC tumors were 5.79 cm ± 3.40 cm and 8.25 cm ± 2.58 cm, respectively, showing a statistically significant difference (P = 0.005). The HCC group exhibited a higher incidence of hepatitis, elevated AFP levels, and elevated prothrombin levels in the HCC group compared to the ICC group. Conversely, elevated CA 19-9 Levels were more common in the ICC group than in the HCC group. These differences were statistically significant (P = 0.007, 0.000, 0.000, and 0.001, respectively). In contrast, no significant difference was observed in the tumor marker CEA between the two groups (P = 0.251), as displayed in Table 1.

Table 1 Clinical data of patients with hepatocellular carcinoma and intrahepatic cholangiocarcinoma.

	Hepatocellular carcinoma (n = 40)	Intrahepatic cholangiocarcinoma (n = 18)	t/χ²	P value
Age	59.95 ± 9.32	61.72 ± 8.17	-0.694	0.490
Sex			12.002	0.001b
Male	35	8
Female	5	10
Size	5.79 ± 3.40	8.25 ± 2.58	-2.891	0.005b
Child-Pugh			2.145	0.143
A	38	15
B	2	3
History of hepatitis			9.908	0.007b
Hepatitis B	25	6
Hepatitis C	5	0
Without	5	12
Laboratory index
Alpha-fetoprotein	32/40	5/18	14.657	0.000b
Carbohydrate antigen 19-9	7/40	11/18	11.031	0.001b
Carcinoembryonic antigen	6/40	5/18	1.319	0.251
Abnormal prothrombin	32/40	3/18	20.808	0.000b

^aP < 0.05.

^bP < 0.01.

Quantitative variables are expressed as means ± SD. Qualitative variables are expressed as raw numbers; numbers in parentheses are proportions followed by percentages.

Interobserver reproducibility

Bland-Altman diagrams of SADC, D, f, Dstar, MK, and MD were used to determine the absolute variability between the two radiologists. The mean difference for SADC was 0.05, with a 95%CI: -0.02 to 0.1, the mean difference for D was 0.02, with a 95%CI: -0.001 to 0.05, the mean difference for f was -13.74, with a 95%CI: -28.25 to 0.78, the mean difference for Dstar was -12.14, with a 95%CI: -22.26 to -2.02, the mean difference for MK was 5.17, with a 95%CI: -20.61 to 30.96, and the mean difference for MD was -0.02, with a 95%CI: -0.07 to 0.03. Among the 66 points, 2 (SADC), 4 (D), 3 (f), 5 (Dstar), 4 (MK), and 5 (MD) points lied outside the 95%CI range, with proportions of 3.00%, 6.06%, 4.55%, 7.58%, 6.06%, and 7.58%, respectively. Only Dstar showed a statistically significant difference (P = 0.020), while no statistically significant differences were found in SADC (P = 0.133), D (P = 0.058), f (P = 0.063), MK (P = 0.690), and MD (P = 0.345) (Figure 1).

Figure 1 Bland-Altman analysis of differences between two radiologists in determining the standard apparent diffusion coefficient, imaging intravoxel incoherent motion (pure diffusion coefficient, perfusion fraction, pseudo diffusion coefficient), and diffusion kurtosis imaging (mean kurtosis coefficient, mean diffusion coefficient) parameters with hepatocellular carcinoma and intrahepatic cholangiocarcinoma. The mean of methods axis represents the geometric mean of both methods. The difference between methods axis represents the difference between both methods. The upper and lower brown dashed horizontal lines indicate the upper and lower limits of the 95% consistency limit; the blue solid horizontal line in the middle represents the average value of the difference, and the green dotted line indicates the 95%CI of the average value of the difference. Dstar: Pseudo diffusion coefficient; MD: Mean diffusion coefficient; MK: Mean kurtosis coefficient; SADC: Standard apparent diffusion coefficient.

Comparison of SADC, IVIM (D), and DKI (MK, MD) parameters

The means ± SD of SADC, IVIM (D), and DKI (MD) parameters in the HCC group were (1) 1.14 × 10⁻³ mm²/second ± 0.23 × 10⁻³ mm²/second; (2) 1.23 × 10⁻³ mm²/second ± 0.24 × 10⁻³ mm²/second; and (3) 1.53 × 10⁻³ mm²/second ± 0.39 × 10⁻³ mm²/second, respectively. These values were significantly lower than the corresponding values of the ICC group: (1) 1.54× 10⁻³ mm²/second ± 0.27 × 10⁻³ mm²/second; (2) 1.52 × 10⁻³ mm²/second ± 0.38 × 10−3 mm²/second; and (3) 1.92 × 10⁻³ mm²/second ± 0.37 × 10⁻³ mm²/second (P < 0.05). The means ± SD of IVIM (f, Dstar) and DKI (MK) parameters in the HCC group were (1) 190.45 × 10⁻³ mm²/second ± 121.71 × 10⁻³ mm²/second; (2) 87.15 × 10⁻³ mm²/second ± 55.83 × 10⁻³ mm²/second; and (3) 706.48 × 10⁻³ mm²/second ± 204.62 × 10⁻³ mm²/second, respectively. These results were higher than those in the ICC group: (1) 167.05 × 10⁻³ mm²/second ± 89.24 × 10⁻³ mm²/second; (2) 49.90 × 10⁻³ mm²/second ± 11.15 × 10⁻³ mm²/second; and (3) 605.19 × 10⁻³ mm²/second ± 82.23 × 10⁻³ mm²/second, respectively. However, only MK showed a statistically significant difference (P < 0.05) (Table 2, Figures 2 and 3).

Figure 2 Surgically confirmed moderately differentiated hepatocellular carcinoma in a 65-year-old man. A: T2-weighted imaging image; B: Diffusion-weighted image with b = 800 mm²/second; C: Pure diffusion coefficient (D) map; D: Perfusion fraction (f) map; E: Standard apparent diffusion coefficient (SADC) map; F: Pseudo diffusion coefficient (Dstar) map; G: Mean kurtosis coefficient (MK) map; H: Mean diffusion coefficient (MD) map. The mean D, f, SADC, Dstar, MK, and MD values for the tumor were 0997 × 10^-3 mm²/second, 92.62 × 10^-3 mm²/second, 1.037 × 10^-3 mm²/second, 55.18 × 10^-3 mm²/second, 651 × 10^-3 mm²/second, and 1.179 × 10^-3 mm²/second, respectively.

Figure 3 Puncture biopsy confirmed intrahepatic cholangiocarcinoma in a 52-year-old man. A: T2-weighted imaging image; B: Diffusion-weighted image with b = 800 mm²/second; C: Pure diffusion coefficient (D) map; D: Perfusion fraction (f) map; E: Standard apparent diffusion coefficient (SADC) map; F: Pseudo diffusion coefficient (Dstar) map; G: Mean kurtosis coefficient (MK) map; H: Mean diffusion coefficient (MD) map. The mean D, f, SADC, Dstar, MK, and MD values for the tumor were 1609 × 10^-3 mm²/second, 143.89 × 10^-3 mm²/second, 1.630 × 10^-3 mm²/second, 134.59 × 10^-3 mm²/second, 552.6 × 10^-3 mm²/second, 2.064 × 10^-3 mm²/second, respectively.

Table 2 Comparison of standard apparent diffusion coefficient, imaging intravoxel incoherent motion (pure diffusion coefficient, perfusion fraction, pseudo diffusion coefficient), and diffusion kurtosis imaging (mean diffusion coefficient and mean kurtosis coefficient) parameters between the hepatocellular carcinoma and intrahepatic cholangiocarcinoma groups.

	Hepatocellular carcinoma	Intrahepatic cholangiocarcinoma	t value	P value
Standard apparent diffusion coefficient (× 10-3 mm²/second)	1.14 ± 0.23	1.54 ± 0.27	-6.073	0.000b
Pure diffusion coefficient (× 10-3 mm²/second)	1.23 ± 0.24	1.52 ± 0.38	-3.70	0.000b
Perfusion fraction (× 10-3)	190.45 ± 121.71	167.05 ± 89.24	0.773	0.443
Pseudo diffusion coefficient (× 10-3 mm²/second)	87.15 ± 55.83	49.90 ± 11.15	1.405	0.165
Mean kurtosis coefficient (× 10-3 mm²/second)	706.48 ± 204.62	605.19 ± 82.23	2.133	0.037a
Mean diffusion coefficient (× 10-3 mm²/second)	1.53 ± 0.39	1.92 ± 0.37	0.686	0.000b

^aP < 0.05.

^bP < 0.01.

Diagnostic performance of SADC, IVIM (D), and DKI (MK, MD)

Based on the positive results in Table 2, the ROC curves for the four parameters SADC, D, MD, and MK were drawn. The single-parameter ROC analyses indicated that the optimal cutoff levels were 1.25 × 10⁻³ mm²/second for SADC, 1.32 × 10⁻³ mm²/second for D, 1.41 × 10⁻³ mm²/second for MD, and 650.2 × 10⁻³ mm²/second for MK. The corresponding AUC were 0.874 (95%CI: 0.676-0.883), 0.793 (95%CI: 0.676-0.883), 0.757 (95%CI: 0.635-0.854), and 0.733 (95%CI: 0.609-0.834), respectively. Moreover, the sensitivities and specificities were 95% and 67.93% for SADC, 80% and 69.57% for D, 100% and 43.48% for MD, and 90% and 67.93% for MK. The Youden indices were 0.6239, 0.4957, 0.4348, and 0.5739, respectively. All valid parameters were combined, yielding an optimal cutoff level of 0.46 × 10⁻³ mm²/second and a joint AUC of approximately 0.895 (95%CI: 0.794-0.957). The sensitivity, specificity, and Youden index were 70%, 93.38%, and 0.6348, respectively (Table 3, Figure 4).

Figure 4 Receiver operating characteristic curves of the standard apparent diffusion coefficient, pure diffusion coefficient, mean kurtosis coefficient, and mean diffusion coefficient for differentiating between hepatocellular carcinoma and intrahepatic cholangiocarcinoma. D: Pure diffusion coefficient; MD: Mean diffusion coefficient; MK: Mean kurtosis coefficient; SADC: Standard apparent diffusion coefficient.

Table 3 Receiver operating characteristic analysis of standard apparent diffusion coefficient, pure diffusion coefficient, mean diffusion coefficient and mean kurtosis coefficient for the prediction and diagnostic performance in the hepatocellular carcinoma and intrahepatic cholangiocarcinoma groups.

	Area under the receiver operating characteristic curve	Cutoff	Youden index	Sensitivity	Specificity	95%CI
Standard apparent diffusion coefficient	0.874	1.25	0.6239	95	67.39	0.769–0.943
Pure diffusion coefficient	0.793	1.32	0.4957	80	69.57	0.676–0.883
Mean kurtosis coefficient	0.733	650.2	0.5739	90	67.39	0.609–0.834
Mean diffusion coefficient	0.757	1.41	0.4348	100	43.48	0.635–0.854
All	0.895	0.46	0.6348	70	93.38	0.794–0.957

DISCUSSION

Our results indicated a larger proportion of male patients in the HCC group compared with the ICC group; in addition, tumors in the HCC group were relatively shorter in length. HCC patients were associated with an increased risk of hepatitis, elevated AFP levels, and elevated prothrombin levels, whereas ICC was associated with a higher proportion of CA 19-9 positivity. Furthermore, the D, SADC, and MD values in the HCC group were higher than those in the ICC group, while the MK value was lower. Subsequently, the parameters derived from SADC, IVIM, and DKI were used for the differential diagnosis of HCC and ICC.

Nonetheless, in the analysis of clinical data, the results regarding gender and tumor length of this study were inconsistent with those of previous studies[24,28]. Despite the higher incidence of HCC in men compared to women, this study included only 18 cases of ICC. Therefore, future comparisons will be made after expanding the sample size. In the present study, 35 out of 40 HCC cases were surgically removed, while only 6 out of 18 ICC cases were surgically removed. Most of the ICC cases showed large masses with a tendency to fuse. These lesions were unsuitable for surgical removal due to their size, shape, and relationship with surrounding tissues. If only cases suitable for surgical removal were included in the study, such errors would be reduced.

IVIM utilizes DWI images with multiple b-values and employs a double exponential model to extract quantitative information, reflecting the diffusion of water molecules and microcirculation perfusion in local tissues[10,29]. In this study, the values of D and SADC showed statistical significance in differentiating ICC from HCC, with higher values observed in ICC compared to HCC. Histologically, a higher number of tumor cells was observed in HCC compared to ICC. In addition, the tumor cells in HCC demonstrated a closer arrangement, leading to blocked extracellular space and limited diffusion coefficients for SADC and D, thereby reducing their values. In contrast, the typical pathological features of ICC include a large amount of mucus matrix and fibrosis in the center, surrounded by highly cellular and vascularized tumor cells[30]. Theoretically, such an arrangement involves a relatively low number of cells in ICC, leading to an increase in SADC and D values[31]. The research results of Shao et al[24] and Wei et al[26] are consistent with our findings. Zhou et al[11] investigated the IVIM results of 70 patients with HCC and revealed higher D and SADC values of well-differentiated HCC compared to those of poorly differentiated HCC. These results suggest that poorly differentiated tumors exhibit rapid proliferation, increasing the number of tumor cells and tumor density while decreasing the amount of intercellular substances, resulting in lower D and SADC values. Ichikawa et al[32] employed IVIM to distinguish between benign and malignant liver tumors, revealing that the D value of malignant tumors was lower than that of benign tumors. Moreover, Yue et al[13] evaluated the IVIM images of 50 Lesions after transcatheter arterial chemoembolization (TACE) and found that the D value of the tumor active area was lower than that of normal liver parenchyma and even lower than that of the tumor necrosis area. These findings indicate that the number of tumor cells is negatively correlated to the D and SADC values.

In this study, the Dstar and f values of HCC were higher than those of ICC but the difference did not reach statistical significance, which is consistent with a previous study[26]. However, Wang et al[23] and Peng et al[25] reported opposite results, while Shao et al[24] suggested that f was significant, but Dstar was not. Both Dstar and f reflect the distribution of blood vessels in tissues. Dstar is related to tissue capillary perfusion, while f is related to the fraction of microvascular perfusion in the total diffusion effect. Theoretically, these two parameters should be distinguishable. The controversial results are primarily attributed to the following reasons. Firstly, the histopathological grades of HCC and ICC included in this study were uneven, with various degrees of tumor differentiation. Therefore, the statistical analysis did not represent a one-to-one comparison of the corresponding pathological grades. The values of Dstar and f mainly depend on the distribution and number of blood vessels. Poorly-differentiated tumors have high perfusion as they contain rich vasculature, whereas highly-differentiated tumors are the opposite[33,34]. Secondly, HCC can be divided into two types based on the vasculature, namely poor blood supply and high blood supply. The Dstar and f values of lesions with rich blood supply are higher than those with poor blood supply[35] In this study, some HCC cases showed poor blood supply, with no obvious enhancement or only slight enhancement in the hepatic artery phase, leading to relatively low Dstar and f values. Finally, microcirculatory perfusion represented by Dstar and f may also be affected by physiological activities such as glandular secretion and blood flow in glandular ducts and catheters[36].

To our knowledge, this is the first study to demonstrate the diagnostic value of DKI parameters in differentiating HCC and ICC. The MK value refers to the average dispersion kurtosis of tissues in all spatial directions, primarily reflecting the complexity and heterogeneity of tissues. MD represents a non-Gaussian distribution with a modified average SADC, mainly reflecting the diffusion of water molecules in tissues. This diffusion is limited by pathological changes leading to high cell density and reduced extracellular space. The cancer cells in HCC exhibit nuclear atypia, and tumor blood vessels proliferate in a distorted manner. Histologically, irregular trabecular growth patterns are observed, as well as solid and pseudo-glandular structures. Cytologically, tumor cells mimic hepatocytes, usually exhibiting transparent (glycogen-rich) or fatty cytoplasm, with complex and diverse components[30]. The center of ICC tumors has extensive necrosis, with tumor components located around the periphery. A lower number of tumor cells is found in ICC compared to HCC, reflecting tumor cell heterogeneity. Additionally, in this study, the ROI was marked on the largest plane. Although large-scale necrosis was avoided, some necrotic areas were inevitably included, which may have reduced some ICC MK values. Goshima et al[15] reported that the MK value of the active part of HCC was lower than that of normal liver parenchyma but higher than that of the necrotic part. In normal liver parenchyma, hepatocytes are arranged in an orderly manner, including diffusion barriers such as fibrous septa and hepatic sinusoids, while containing only a small number of completely necrotic HCC cells, usually developing into coagulation necrosis. Yuan et al[14] used MK to evaluate the therapeutic effect of tumors after radiofrequency ablation (RFA) and found that the MK of completely necrotic HCC decreased significantly after RAF. Wu et al[37] reported that the degree of malignancy of HCC was positively correlated with the MK value and negatively correlated with the MD value. Furthermore, Budjan et al[38] showed that the MD value of HCC was significantly lower than that of common benign tumors in the liver. Jia et al[39] conducted similar research and revealed a higher MK value in HCC compared to benign liver tumors, whereas the MD value was lower than that of benign masses. The MK value of MVI in HCC was greater than that of MVI-negative HCC. MVI-positive HCC is characterized by tumor embolism or increased cancer cells in venous branches, with tumor cells diffusing and infiltrating through the microcirculation. The presence of MVI leads to a more complex microstructure composition in the tumor[9,12]. MK is widely used clinically to reflect the complexity of tissues. The higher the MK value, the higher the tumor cell activity, pathological grade, and positive rate of MVI. The increase in MD value was negatively correlated with the above factors, which is consistent with our research results.

SADC is a non-specific parameter influenced by both microcirculation-related perfusion and cell-related diffusion. Contrary to previous research[23-26], the differential diagnosis efficiency of SADC for HCC and ICC was found to be higher than that of D. The above 4 studies included one or two b-values greater than 800, but this case did not choose, 5 or 7 Low b-values were selected, and six were chosen in this study. At low b-values, the attenuation of the diffusion-weighted MR signal is mainly driven by Dstar, while at high b-values, it is primarily driven by D[40]. Many studies have confirmed that the selection and number of b-values affect IVIM parameters[40-43]. No uniform standard has been established for the selection of b-values globally. Multiple b-value selection can improve result accuracy, which also increases scanning time. Therefore, optimizing the number and distribution of b-values plays an essential role in reducing the error in IVIM parameter measurement. Regarding the diagnostic efficiency of SADC, the scholars[21,44] found that SADC was more accurate than D in evaluating the changes 24-48 hours after TACE, distinguishing HCC with different pathological differentiation and identifying the expression of cytokeratin 19 (CK 19) in HCC. This study also demonstrates the high diagnostic efficiency of SADC, and its clinical application value needs further exploration. In our results, the AUC of MD and MK were similar and close. Currently, DKI[12,14,15] overlap with IVIM in routine clinical applications[18,20,22]. They are only used in combination in the comparative analysis of CK 19 expression in HCC[21]. The diagnostic efficiency of MD in DKI is higher than that of SADC, D, and Dstar in IVIM. Compared with previous research results, the AUC of our SADC and D was close to or greater than most of the above findings, and the diagnostic efficiency reached its maximum after DKI fusion. Their combined application will provide more evidence for solving clinical problems in the future.

Nevertheless, the limitations of the present study should be acknowledged. First, this study is a retrospective case-control study that inevitably involves selection bias. Second, although the number of cases in this study is relatively small, we have achieved valuable results. We look forward to further research with larger datasets. Third, the pathological differentiation of included HCC and ICC cases was uneven, and the tumors could not be compared to those with a corresponding classification or degree of differentiation. Fourth, the delineation of the ROI was performed on the largest slice of the tumor, whereas measuring the entire solid tumor volume could yield higher diagnostic performance[45].

CONCLUSION

This study demonstrated that SADC, D, MK, and MD can be used as reliable and stable parameters for the differential diagnosis of HCC and ICC. The combined model showed significantly improved diagnostic efficiency.