Nomogram based on the neutrophil-to-lymphocyte ratio and MR diffusion quantitative parameters for predicting Ki67 expression in hepatocellular carcinoma from a prospective study

Abstract

This study aimed to establish and validate a multiparameter prediction model for Ki67 expression in hepatocellular carcinoma (HCC) patients while also exploring its potential to predict the one-year recurrence risk. The clinical, pathological, and imaging data of 83 patients with HCC confirmed by postoperative pathology were analyzed, and the patients were randomly divided into a training set (n = 58) and a validation set (n = 25) at a ratio of 7:3. All patients underwent a magnetic resonance imaging (MRI) scan that included multi-b value diffusion-weighted scanning before surgery, and quantitative parameters were obtained via intravoxel incoherent motion (IVIM) and diffusion kurtosis (DKI) models. Univariate and multivariate logistic regression analyses were conducted using the training set data to construct a model, which was internally validated. The area under the curve (AUC) of the receiver operating characteristics (ROC), a decision curve analysis (DCA), and a calibration analysis were used to evaluate the model’s performance. Additionally, for patients with available follow-up data, the combined model was evaluated for its potential utility in predicting the one-year recurrence risk by analyzing the area under the curve (AUC) of the receiver operating characteristic (ROC) curve.The combined model outperformed the clinicaland parametric models in predicting high Ki67 expression. The nomograms based on the combined model included the neutrophil-to-lymphocyte ratio (NLR), ADCslow_Aver. The model showed strong discrimination in the training set, with an AUC of 0.836 (95% CI: 0.729–0.942) and acceptable calibration (Hosmer–Lemeshow p = 0.109). In the validation set, the model maintained moderate discrimination (AUC 0.806, 95% CI: 0.621–0.990) with good calibration (p = 0.663). DCA revealed that the combined model provided good clinical value and correction effects. Additionally, when used to predict the one-year recurrence risk, the combined model achieved moderate accuracy (AUC = 0.747), highlighting its potential utility in identifying patients at a higher risk of recurrence. A nomogram incorporating the NLR and quantitative MR diffusion parameters effectively predicts Ki67 expression in HCC patients before surgery. The model also shows promise in predicting recurrence risk, which may aid in postoperative risk stratification and patient management.

Clinical Relevance Statement We established a model that incorporated the NLR and quantitative magnetic resonance diffusion parameters, which demonstrated robust performance in predicting both high Ki67 expression and the one-year recurrence risk in HCC patients. This model shows potential clinical value in guiding postoperative risk stratification and personalized treatment planning.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-024-82333-7.

Introduction

Hepatocellular carcinoma (HCC) is a malignant tumor with the sixth-highest incidence and the third-highest mortality rate worldwide¹, and the prognosis for patients with advanced HCC remains poor. The early and accurate diagnosis of HCC is thus crucial for facilitating personalized treatment, improving the prognosis of patients, and reducing mortality rates². Identifying reliable biological markers can contribute significantly to achieving these goals.

Ki67, a monoclonal antibody expressed mainly in the nucleus, has emerged as a promising prognostic marker for HCC. It serves as an important indicator for measuring the degree of cell proliferation and the biological behavior of cells within the tumor microenvironment^3,4. High Ki67 expression often indicates that HCC has greater malignant potential, serves as an independent risk factor for disease-free and overall survival outcomes of HCC patients, and functions as a reliable prognostic indicator for HCC^5,6. Pathological analysis via needle biopsy or surgical resection provides evidence of Ki67 expression; however, few preoperative noninvasive assessments of Ki67 expression have been performed.

Noninvasive imaging techniques, especially magnetic resonance imaging (MRI), offer potential alternatives. MRI multiparameter diffusion-weighted imaging is an effective noninvasive tool for diagnosing HCC and evaluating its prognosis and treatment efficacy. Some studies have demonstrated that preoperative MRI parameters can predict high and low Ki67 expression in HCC⁷. In the evaluation of the liver cancer microenvironment, the intravoxel incoherent motion (IVIM) and diffusion kurtosis imaging (DKI) models derived from DWI have demonstrated effectiveness^8,9, and they can also be used to evaluate biological behaviors, such as microvascular infiltration in HCC tumors^10–12.

Given the close relationship between inflammation and tumor progression^13,14, thet inflammatory response is another important factor requiring consideration. The neutrophil-to-lymphocyte ratio (NLR), a marker of systemic inflammatory response, has shown potential as a predictor of immune status and survival outcomes in cancer patients¹⁵. The preoperative peripheral blood NLR has been reported to predict the recurrence and survival rates of patients with HCC^15–17. Further research is needed to confirm whether there is a correlation between the NLR, an indicator of inflammatory status, and Ki67, an indicator of tumor cell proliferation.

In this study, we aimed to explore whether imaging and clinical parameters can serve as biological indicators for the preoperative prediction of Ki67 expression in HCC patients. We analyzed the correlation between Ki67 expression and preoperative diffusion-weighted MR imaging parameters and MRI features, as well as the peripheral blood NLR. Additionally, we developed multiple preoperative prediction models for Ki67 expression and evaluated their predictive performance.

Materials and methods

Patients

This study was approved by the Ethics Committee of the First Affiliated Hospital of Guangxi Medical University (Approval Number: 2024-E432-01). We systematically collected data from 354 patients with primary liver cancer who underwent liver MRI scans within 1 month before surgery at the First Affiliated Hospital of Guangxi Medical University from March 2021 to December 2023. All patients signed provided their written informed consent.

The inclusion criteria were as follows: (1) patients who underwent contrast-enhanced MRI of the liver, including IVIM-DKI sequences, within 1 month before surgery and were diagnosed with primary liver cancer; (2) patients with a pathologically confirmed diagnosis of HCC after surgical resection; and (3) patients with a solitary liver tumor with a diameter greater than 2 cm.

The exclusion criteria were as follows: (1) history of preoperative surgical resection, radiotherapy, or chemotherapy; (2) unsatisfactory imaging data; or (3) incomplete clinical or pathological data.

The flowchart illustrating the inclusion and exclusion criteria for this study is shown in Fig. 1. The patients were randomly divided into training and validation sets at a ratio of 7:3 via the R function “createDataPartition” to ensure a balanced distribution of outcome events between these two sets.

Fig. 1.

Flowchart of the inclusion and exclusion criteria of this study.

Clinical Data Collection

The clinical features included age, sex, weight, neutrophil count, absolute lymphocyte count, alpha-fetoprotein (AFP), alanine aminotransferase (ALT), aspartate aminotransferase (AST), and hepatitis B virus (HBV) infection. The pathological features included the degree of tumor differentiation and the percentage of Ki67 expression.

MRI technique

MRI scans were performed on a 3.0T MRI scanner (MAGNETOM Prisma, Siemens Healthineers, Germany) with a 16-channel body coil, and were performed alongside respiratory gating. This technique synchronized image acquisition with the patient’s breathing cycle, helping to minimize motion artifacts that could obscure image quality, especially in areas that moved when the patient breathed, such as the abdomen.

All patients underwent multiphase gadoxetate-ethoxybenzyl-diethylen etriamine pentanoate (Gd-EOB-DTPA) enhanced scanning, and the IVIM-DKI scan sequence was added. The b values used for the IVIM-DKI sequence were 0, 20, 50, 100, 150, 200, 600, 1000, 2000, and 3000 s/mm². The scanning parameters are shown in Table 1.

Table 1.

MRI scanning parameters.

	TR(ms)	TE(ms)	slice thickness (mm)	slice gap (mm)		FOV(mm)		Matrix
T1WI	3.97	129	5.0		2.0		380 × 380		177 × 320
T2WI	1000.0	95.0	5.0		2.0		380 × 380		192 × 192
Hepatobiliary phase	4.48	1.33	3.0		2.0		380 × 380		163 × 288
IVIM-DKI	4900	57	5.0		6.4		380 × 261		88 × 128

Note: TR: Repetition Time; TE: Echo Time; FOV: Field of View; T1WI: T1-weighted image; T2WI: T2-weighted image; IVIM-DKI: intravoxel incoherent motion‒diffusion kurtosis imaging.

Image analysis

Two radiologists with 3 and 5 years of experience in abdominal imaging diagnosis, who were blinded to the clinical and histopathological information, independently analyzed the patients’ MR images and postprocessed the IVIM-DKI images using MR software (Body Diffusion 1.4.0, Siemens Healthineers, Germany). In instances of disagreement, resolution was achieved through consultation with a senior radiologist with 10 years of experience in abdominal imaging diagnosis.

The following MRI findings were assessed as previously described¹⁸: (1) background of liver cirrhosis; (2) maximum tumor diameter; (3) tumor location; (4) tumor margin: observation of whether there are nodular protrusions on the tumor edge; (5) tumor pseudocapsule: thin annular high signal at the edge during the venous phase; (6) peritumoral enhancement: irregular enhancement area of the peritumoral liver parenchyma during the arterial phase; (7) peritumoral hypointensity: irregular low signal area of the peritumoral liver parenchyma during the hepatobiliary phase; and (8) necrosis, cyst, or hemorrhage.

In the DKI model, the b values were 0, 200, 600, 1000, 2000, and 3000 s/mm²; in the IVIM model, they were 0, 20, 50, 100, 150, 200, 600, and 1000 s/mm²; in the conventional single-exponential model, they were 0, 500, and 1000 s/mm². In the ADC model, with ultrahigh b values, they were 2000 and 3000 s/mm².

Three ROIs were drawn along the tumor edge at b = 0 s/mm². IVIM-DKI images, including the maximum cross-sectional position of the tumor and its upper and lower levels, were collected, with blood vessels, the bile duct, hemorrhage, and imaging artifacts avoided by referring to T1WI and T2WI sequences. These images were then copied into different parameter maps. For each ROI fraction, the average value was recorded as MK_Aver, MD_Aver, D_Aver, Dp_Aver, f_Aver, ADCslow_Aver, and ADCuh_Aver.

The mean kurtosis (MK) and mean diffusivity (MD) parameters from the DKI model; the true diffusion coefficient (D), perfusion-dependent diffusion coefficient (Dp), and perfusion fraction (f) parameters from the IVIM model; the ADC value (ADCslow) from the conventional single-exponential model; and the ADC value (ADCuh) from the ultrahigh b value were all obtained.

Follow-up data Collection

Follow-up data were collected for at least one year post-surgery to monitor very early recurrence or death. Very early recurrence was defined as the presence of intrahepatic and/or extrahepatic recurrence within one year following HCC resection¹⁹. The follow-up included imaging examinations (such as CT, MRI, and ultrasound) and serum tumor marker evaluation (such as AFP). These data were obtained from the patients’ routine follow-up visits at the hospital. While every effort was made to gather follow-up data from all patients, some were lost to follow-up, and as a result, complete follow-up data were not available for all individuals. The cutoff date for follow-up was March 31, 2024.

Statistical analysis

All statistical analyses were performed using SPSS (RRID: SCR_002865, version 26.0) and the R Project for Statistical Computing (RRID: SCR_001905, version 4.4.0). The intraclass correlation coefficient (ICC) was used to evaluate the consistency of the two radiologists in observing the same imaging data. When the ICC > 0.75, the consistency was considered to be strong.

The Shapiro‒Wilk test was used to determine whether the measurement data conformed to a normal distribution. Then, Pearson and Spearman rank correlation analyses were performed for normally and nonnormally distributed data, respectively. For the Ki67 low- and high-expression groups, normally distributed data were expressed as the means ± standard deviations (SDs), and they were compared via independent sample t-tests. Nonnormally distributed measurement data were represented by M (P25, P75) and compared via the Mann‒Whitney U test. The chi-square test or Fisher’s exact test was used to compare the differences in enumeration data.

The ROC curve was constructed using the “roc_curve” package, which was used to analyze the predictive performance. The confidence interval for the mean was set to 95%, and when the P value was less than 0.05, the difference was considered significant. The “rms” package was used to plot the calibration curve to evaluate the model’s goodness of fit, and the Hosmer–Lemeshow test was performed.

Results

Patient characteristics

A total of 83 patients, including 65 males and 18 females, were enrolled in this study. In the entire population, as well as in the training and validation sets, 46 (55.4%), 30 (51.7%), and 16 (64%) patients presented with high Ki67 expression (> 10%), respectively. The average ages were 54.39 ± 10.94, 54.52 ± 11.60, and 54.08 ± 9.45 years, respectively, and the average weights were 63.35 ± 10.14, 62.71 ± 11.09, and 64.84 ± 7.49 kg, respectively.

In the assessment of the diffusion-weighted MR imaging parameters, both radiologists demonstrated strong interobserver agreement. The interobserver ICC values of MK_Aver, MD_Aver, f_Aver, D_Aver, Dp_Aver, ADCslow_Aver, and ADCuh_Aver were 0.904, 0.924, 0.908, 0.931, 0.914, 0.893, and 0.906 respectively.

The clinical, pathological, and MRI features, as well as the MRI diffusion-weighted imaging parameters of the enrolled patients, are summarized in Table 2. These indicators were comparable between the training and validation sets (P > 0.05).

Table 2.

Comparison of patient characteristics and parameters.

	Total(n = 83)			Training set(n = 58)
	Validation (n = 25)	Training (n = 58)	Pα	Low KI67 expression group (n = 28)	High KI67 expression group (n = 30)	Pβ
Clinical data
Age(years)	54.08 ± 9.45	54.52 ± 11.60	0.869	56.43 ± 11.10	52.73 ± 11.96	0.229
Sex, n(%)			0.160			0.179
Male	22(88.00%)	43(74.14%)		23(82.14%)	20(66.67%)
Female	3(12.00%)	15(25.86%)		5(17.86%)	10(33.33%)
Weight(kg)	64.84 ± 7.49	62.71 ± 11.09	0.311	65.16 ± 12.00	60.42 ± 9.81	0.104
HBV, n(%)			0.257			0.670
No	4(16.00%)	16(27.59%)		7(25.00%)	9(30.00%)
Yes	21(84.00%)	42(72.41%)		21(75.00%)	21(70.00%)
AFP	20.73 (7.08, 381.88)	55.90 (4.29, 1071.90)	0.972	5.76 (2.54, 87.26)	503.39 (20.11, 16484.02)	< 0.001*
ALT	41.00 (24.00, 53.00)	28.00 (20.00, 41.75)	0.101	30.00 (21.75, 48.25)	23.50 (16.25, 38.75)	0.196
AST	39.00 (29.00, 50.00)	32.00 (26.00, 54.50)	0.659	29.50 (24.75, 47.75)	37.50 (27.25, 64.75)	0.114
NLR	2.16 (1.55, 2.64)	2.11 (1.77, 2.86)	0.497	1.97 (1.68, 2.20)	2.72 (2.02, 3.91)	< 0.001*
Differentiation grade, n(%)			0.149			< 0.001*
High	6(24.00%)	5(8.62%)		5(17.86%)	0(0.00%)
Middle	16(64.00%)	41(70.69%)		22(78.57%)	19(63.33%)
Low	3(12.00%)	12(20.69%)		1(3.57%)	11(36.67%)
MRI imaging features
Background of liver cirrhosis, n(%)			0.276			0.444
No	8(32.00%)	26(44.83%)		14(50.00%)	12(40.00%)
Yes	17(68.00%)	32(55.17%)		14(50.00%)	18(60.00%)
Tumor maximum diameter(cm)	5.66 ± 3.64	5.31 ± 2.87	0.643	4.22 ± 2.38	6.32 ± 2.96	0.005*
Tumor location, n(%)			0.602			0.543
Left lobe	8(32.00%)	14(24.14%)		5(17.86)	9(30.00)
Right lobe	13(52.00%)	37(63.79%)		20(71.43)	17(56.67)
Left and right lobes	4(16.00%)	7(12.07%)		3(10.71)	4(13.33)
Tumor margin, n(%)			0.830			0.037*
Smooth	11(44.00%)	27(46.55%)		17(60.71)	10(33.33)
Rough	14(56.00%)	31(53.45%)		11(39.29)	20(66.67)
Tumor pseudocapsule, n(%)			0.195			0.002*
Complete	8(32.00%)	11(18.97%)		10(35.71)	1(3.33)
No/incomplete	17(68.00%)	47(81.03%)		18(64.29)	29(96.67)
Peritumoral enhancement, n(%)			0.205			0.019*
No	15(60.00%)	26(44.83%)		17(60.71)	9(30.00)
Yes	10(40.00%)	32(55.17%)		11(39.29)	21(70.00)
Peritumoral hypointensity, n(%)			0.276			0.003*
No	17(68.00%)	32(55.17%)		21(75.00)	11(36.67)
Yes	8(32.00%)	26(44.83%)		7(25.00)	19(63.33)
Necrosis, cyst, or hemorrhage, n(%)			0.170			0.036*
No	6(24.00%)	23(39.66%)		15(53.57)	8(26.67)
Yes	19(76.00%)	35(60.34%)		13(46.43)	22(73.33)
MRI quantitative parameters
MK_Aver (×10−6 mm2/s)	690.94 ± 101.55	669.70 ± 91.51	0.351	633.89 ± 79.58	703.13 ± 90.38	0.003*
MD_Aver (×10−6 mm2/s)	1643.33 (1438.59,2006.15)	1594.99 (1419.16, 1899.26)	0.747	1769.11 (1504.32, 2282.08)	1521.22 (1275.33, 1753.47)	0.026*
D_Aver (×10−6 mm2/s)	903.95 ± 253.18	858.21 ± 212.68	0.399	912.32 ± 231.61	807.71 ± 183.02	0.061
f_Aver (×10−3)	236.21 (157.73, 305.95)	206.70 (169.01, 312.56)	0.886	220.25 (179.85, 332.17)	200.51 (160.98, 260.53)	0.318
Dp_Aver (×10−6 mm2/s)	153.77 (110.69, 224.91)	183.28 (152.47, 232.41)	0.133	179.01 (153.40, 228.37)	190.19 (155.39, 242.93)	0.521
ADCslow_Aver (×10−6 mm2/s)	1139.46 (953.21, 1330.27)	1156.25 (1027.16, 1305.79)	0.516	1277.74 (1198.13, 1365.38)	1038.21 (938.93, 1150.22)	< 0.001*
ADCuh_Aver (×10−6 mm2/s)	478.18 (422.81,551.49)	478.83 (443.29, 528.75)	0.897	478.83 (455.33, 529.64)	477.95 (436.88, 525.38)	0.799

HBV: Hepatitis B virus; AFP: Alpha-fetoprotein; ALT: Alanine aminotransferase; AST: Aspartate aminotransferase; NLR: Neutrophil-to-lymphocyte ratio; MRI: Magnetic resonance imaging; MK: Mean kurtosis; MD: Mean diffusivity; D: True diffusion coefficient; f: Perfusion fraction; Dp: Perfusion-related diffusion coefficient; ADCslow: Slow apparent diffusion coefficient; ADCuh: Apparent diffusion coefficient under ultra-high b values; t: t test; Z: Mann‒Whitney test; χ²: chi-square test; Pα: comparison between the training group and the validation group; Pβ: comparison between the Ki67 high- and low-expression groups in the training group. *indicates P < 0.05.

Model development

>A total of 24 variables were analyzed. As shown in Tables 2 and 12 indicators significantly differed between the Ki67 high- and low-expression groups in the training set (P < 0.05). Given the significant difference in proportion between different degrees of pathological differentiation, we included only the remaining 11 indicators in the univariate logistic regression.

As shown in Table 3, nine important variables related to high Ki67 expression in HCC were identified, including the NLR (P < 0.01), the maximum tumor diameter (P < 0.01), rough margin (P = 0.04), absent or incomplete tumor pseudocapsule (P = 0.01), peritumoral enhancement (P = 0.02), peritumoral hypointensity (P < 0.01), necrosis, cyst, or hemorrhage (P = 0.04), MK average (P < 0.01), and ADCslow_Aver (P < 0.01). Incorporating these indicators into multivariate logistic regression analysis revealed that the NLR and ADCslow_Aver were independent risk factors for high Ki67 expression in HCC patients. The NLR was used to construct a clinical model for high Ki67 expression in HCC, while ADCslow_Aver was used to construct the parametric model.The NLR and ADCslow_Aver were used to develop the combined model.

Table 3.

Results of the univariate and multivariate logistic regression analyses of each variable in the training set note: AFP: Alpha-fetoprotein; NLR: neutrophil-to-lymphocyte ratio; MRI: magnetic resonance imaging; MK: Mean kurtosis; MD: Mean diffusivity; ADCslow: slow apparent diffusion coefficient; OR: odds ratio; CI: confidence interval; * indicates P < 0.05.

	Univariate logistic regression		Multivariate logistic regression
	P	OR (95%CI)	P	OR (95%CI)
Clinical data
AFP	0.10	1.00 (1.00–1.00)
NLR	< 0.01*	3.72 (1.59–8.70)	0.03*	3.84 (1.12–13.14)
Imaging features of MRI
Tumor maximum diameter	< 0.01*	1.36 (1.08–1.71)	0.96	0.99 (0.63–1.56)
Tumor margin	0.04*	3.09 (1.06–9.04)	0.19	3.17 (0.57–17.61)
Tumor pseudocapsule	0.01*	16.11 (1.90–136.62)	0.29	4.11 (0.31–55.34)
Peritumoral enhancement	0.02*	3.61 (1.21–10.71)	0.44	0.41 (0.04–4.02)
Peritumoral hypointensity	< 0.01*	5.18 (1.67–16.09)	0.12	5.93 (0.63–56.10)
Necrosis, cyst, or hemorrhage	0.04*	3.17 (1.06–9.52)	0.75	0.73 (0.11–5.05)
MRI quantitative parameters
MK Aver	< 0.01*	1.01 (1.01–1.02)	0.30	0.99 (0.98–1.01)
MD Aver	0.06	1.00 (1.00–1.00)
ADCslow_Aver	< 0.01*	0.99 (0.99–0.99)	0.04*	0.99 (0.99–0.99)

Comparison of the predictive performance of different models

The effectiveness of the three models in predicting high Ki67 expression is presented in Table 4; Fig. 2. The combined model outperformed the standalone clinical and parametric in predicting high Ki67 expression, with higher AUCs observed in the training and validation sets (0.836 and 0.806, respectively).

Table 4.

ROC curve analysis for prediction of high and low Ki67 expression using different combined models.

	Training set			Validation set
	AUC (95%CI)	Sensitivity(%)	Specificity(%)	AUC (95%CI)	Sensitivity(%)	Specificity(%)
Clinical model	0.751 (0.623–0.880)	82.1%	63.3%	0.771(0.556–0.986)	77.8%	43.8%
Parametric model	0.817 (0.704–0.929)	85.7%	70.0%	0.764(0.562–0.966)	88.9%	68.8%
Combined model	0.836(0.729–0.942)	96.4%	63.3%	0.806(0.621–0.990)	77.8%	81.2%

Magnetic resonance imaging; AUC: area under the curve; CI: confidence interval.

Fig. 2.

ROC curves of each model in the training (A) and validation (B) sets.

As shown in Fig. 3, the combination model exhibited a high net gain in clinical decision-making. The calibration curve of each model demonstrated that the predicted probability was almost identical to the actual probability in the training and validation sets (Fig. 4).

Fig. 3.

Clinical decision analysis of each model in the training (A) and validation (B) sets. The abscissa is the threshold probability and the ordinate is the net benefit. The gray curve indicates that all patients received the intervention, while the black horizontal line indicates that if the patients receive no intervention, the net benefit would be zero.

Fig. 4.

Calibration curves of the models in the training (A) and validation (B) sets. The dashed line represents the ideal predictive performance, while the solid line represents the model’s predictive performance. The closer the solid line is to the dotted line, the better the model’s prediction accuracy.

Nomogram establishment

A risk prediction model was constructed using a nomogram based on the combined model (Fig. 5). The points of each variable were added to obtain the total points. The total points correspond to the probability of high Ki67 expression. A Hosmer–Lemeshow test was conducted on the model, and the P values for the training and validation sets were 0.109 and 0.663, respectively, both of which are > 0.05. This finding further indicates that the model exhibits good fit.

Fig. 5.

Construction of a nomogram based on the combined model. Note: MK: Mean kurtosis; ADCslow: Slow apparent diffusion coefficient; NLR: Neutrophil-to-lymphocyte ratio.

One-year recurrence analysis

Ultimately, one-year follow-up data were collected for 69 patients.Among them, 35 patients presented with high Ki67 expression(12 cases experienced recurrence), while 34 patients presented with low Ki67 expression(7 cases experienced recurrence). As shown in Fig. 6A, the one-year recurrence-free survival rate was higher in the low Ki67 expression group (79.4%) than in the high Ki67 expression group (65.7%). However, the Log-rank test indicated that this difference was not statistically significant (P = 0.179), possibly due to the small sample size, which could have limited the ability to detect significant differences between the groups.

Fig. 6.

(A) One-year recurrence-free survival curves for patients with high and low Ki67 expression. (B) ROC curve of the combined model for predicting one-year recurrence.

Figure 6B showed the ROC curve of the combined model for predicting one-year recurrence, with an AUC of 0.747 (95% CI: 0.601–0.894), indicating meaningful predictive capability. This model not only predicted high Ki67 expression but also effectively forecasted one-year recurrence, highlighting its clinical utility for early identification and intervention of high-risk patients.

Typical case image

Figure 7 shows the typical MR and pathological images of a patient.

Fig. 7.

A 42-year-old male patient with poorly differentiated hepatocellular carcinoma with high Ki67 expression. (A) Transverse T2WI of the tumor, with high signal intensity and intratumoral patchy and spotty higher signal foci. (B–E) At the arterial phase (B), the tumor exhibited nonuniform enhancement with intratumoral patches of nonenhanced necrosis, whereas the peripheral liver parenchyma showed marked patchy enhancement. During the portal venous (C) and transitional phases (D), the tumor demonstrated diminished enhancement; in contrast, during the hepatic parenchymal phase (E), it exhibited low signal intensity. The peripheral tumor exhibited slightly lower patchy signal intensity. Additionally, there is an incomplete pseudocapsule around the tumor with irregular nodular edges. (F–N) Original intravoxel incoherent motion‒diffusion kurtosis imaging (IVIM-DKI) images at b = 0 s/mm², the MK, MD, f, D, Dp, conventional ADC, and high b value ADC maps, and the histological examination via light microscopy (HE×100). Note: T2WI: T2 weighted image; MK: Mean kurtosis; MD: Mean diffusivity; f: Perfusion fraction; D: True diffusion coefficient; Dp: Perfusion-related diffusion coefficient; ADC: Apparent diffusion coefficient; HE: Hematoxylin-eosin.

Discussion

We established a model incorporating the NLR and ADCslow_Aver that demonstrated good performance in predicting high Ki67 expression. Additionally, this model showed the ability to predict recurrence within one year, further indicating its potential utility in the clinic. By providing valuable insights into both proliferation and recurrence risks, this model may serve as a useful tool for identifying high-risk HCC patients and guiding clinical decision-making.

The NLR reflects the immune status of cancer patients²⁰, with elevated NLR values significantly associated with risk factors for tumor malignancy. Our study revealed that a high NLR was associated with increased Ki67 expression in tumors, indicating poor patient prognosis. Our findings align with previously reported findings^21–23. Ki67 is related to tumor cell proliferation, and the associations of the NLR with tumor angiogenesis, immune evasion, and metastatic diseases are linked to poor prognosis in HCC patients^24,25, which may explain why the NLR and Ki67 expression are highly correlated. This association can be attributed to the proinflammatory environment produced by chronic liver disease, which leads to inflammatory neutrophilism that inhibits the cytolytic activity of other immune cells, such as lymphocytes and activated T cells, thereby suppressing tumor cell apoptosis²⁶. Additionally, neutrophilia can stimulate tumor neovascularization through the secretion of vascular endothelial growth factor (VEGF)^27,28.

We also analyzed the role of other clinical and pathological factors. The pathological grade of HCC can reflect the biological behavior of the tumor and aid in assessing postoperative recurrence²⁹. A previous study³⁰ revealed that the higher the expression of Ki67 is, the greater the degree of tumor invasion and heterogeneity, and the lower the degree of pathological differentiation. Our results were similar; however, we need to increase the sample size for each degree of differentiation for further confirmation. AFP is an independent risk factor for HCC tumor differentiation and patient survival³¹. Wu et al.³² demonstrated that AFP correlated with Ki67 expression in HCC, which aligned with our findings. However, the AFP level was not significant in the univariate logistic regression, which warrants further investigation. ALT and AST are important liver serum enzymes, and they serve as sensitive indicators for assessing impaired liver function³³. This study revealed no significant difference in ALT or AST between the high- and low-expression groups of Ki67, possibly attributable to the lack of significant disparity in HBV infection and cirrhosis cases between the two groups.

Certain imaging features showed P values < 0.05 in univariate logistic regression, indicating a potential association with Ki67 expression, such as tumor pseudocapsule, tumor margin, and peritumoral hypointensity.The dense tumor pseudocapsule restricts tumor cells to the tumor boundary, and its narrow vessels prevent cancer cells from passing through³⁴. When the pseudocapsule is incomplete or absent, HCC tends to exhibit infiltrative growth. Nonsmooth tumor margins³⁵ and peritumoral hypointensity³⁶ in the hepatobiliary phase are independent predictors of microvascular invasion (MVI) in HCC patients and have a significant impact on tumor dissemination and aggressive tumor behavior³⁴. These imaging features correlate with the Ki67 index, possibly because they facilitate tumor growth and invasion.

Diverse perspectives exist regarding the correlation between multiple-parameter diffusion-weighted MR imaging parameters and Ki67 expression. The MK and MD values were not correlated with Ki67 expression in a study conducted by Yuan³⁷. However, Zhang³⁸ reported that the Ki67 high-expression group had lower standard ADC, D, and MD values and high MK values, which aligned with our results. This effect may be attributed to the fact that tumors in the high Ki67 expression group possess a more intricate tissue architecture and exhibit a richer blood supply than those in the low Ki67 expression group. Therefore, the degree to which water molecules inside the tumor deviate from the Gaussian distribution and move in a disordered manner, as well as the degree of tumor blood flow, are greater³⁹. The MK value is an indicator of organizational complexity, whereas the MD value is the average diffusion degree of water molecules in a diffusion gradient field, which can reflect the overall diffusion situation and is independent of direction^11,40. Our results were consistent with those of previous studies, which showed that ADCslow eliminated the effect of microcirculation and could better reflect water restriction due to tumor density and changes in extracellular volume or deposition during tumor proliferation³⁹, which was highly correlated with Ki67 expression, indicating a tumor cell proliferation-related diffusion restriction effect⁴¹.

To the best of our knowledge, this study is the first to systematically combine MRI diffusion parameters and NLR to predict Ki67 expression, offering a valuable contribution to current noninvasive predictive methods for HCC. Previous studies have often been limited to MRI features or parameters alone^7,42, which can be subjective and lack a comprehensive biological context.By combining MRI diffusion parameters with the inflammatory marker NLR, this approach not only improves the accuracy of tumor proliferation prediction, but also provides insights into the inflammatory tumor microenvironment, offering a novel perspective on tumor behavior in HCC.

This study has several limitations. As a prospective study; although we scanned many patients, only a small number of enrolled patients met the inclusion and exclusion criteria, resulting in a relatively modest sample size that could have affected the model’s generalizability. Additionally, as a single-center study, it lacked external validation; therefore, further studies in diverse clinical settings would be beneficial. The study also included a limited number of variables, which may have restricted the model’s predictive power and comprehensive analysis. We also currently lack long-term follow-up data that would allow us to fully assess the model’s prognostic value over time. Future efforts will focus on increasing the sample size and incorporating more comprehensive clinical and imaging features to develop a more robust prediction model and to comprehensively evaluate its long-term clinical value.

Conclusion

The nomogram model we constructed uses two selected factors, including the NLR and MR diffusion quantitative parameters. This model demonstrates robust predictive power and serves as a valuable clinical decision-making tool with favorable net benefits and calibration.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Abbreviations

ADC

Apparent diffusion coefficient

ADCslow

Slow apparent diffusion coefficient

ADCuh

Apparent diffusion coefficient under ultra-high b values

AFP

Alpha-fetoprotein

ALT

Alanine aminotransferase

AST

Aspartate aminotransferase

AUC

Area under the curve

CI

Confidence interval

D

True diffusion coefficient

DCA

Decision curve analysis

Dp

Perfusion-related diffusion coefficient

DKI

Diffusion kurtosis imaging

DWI

Diffusion weighted imaging

f

Perfusion fraction

Gd-EOB-DTPA

Gadoxetate-ethoxybenzyl-diethylen etriamine pentaace

HBV

Hepatitis B virus

HCC

Hepatocellular carcinoma

ICC

Intraclass correlation coefficient

IVIM

Intravoxel incoherent motion

MVI

Microvascular invasion

MK

Mean kurtosis

MD

Mean diffusivity

MRI

Magnetic resonance imaging

NLR

Neutrophil-to-lymphocyte ratio

ROI

Region of interest

ROC

Receiver operating characteristic

RRID

Research resource identifier

Author contributions

Yu-chen Wei and Liang yun: Performed the experiments, Contributed to the design of the studyAnalyzed the data , Wrote the paper.Yan-ling Liang: Collected and assembled data, Contributed to the writing of the manuscript.Contributed to the design of the study.Robert Grimm, PhD done software.Chongze Yang and Yuan-fang Tao: Performed data analysis.Provided critical feedback and helped shape the research.Sheng-chen Jiang: Performed data analysis, Contributed to the writing of the manuscript.Jin-yuan Liao: Corresponding author, Provided funding, Supervised the research, Final approval of the version to be published.

Funding

This work was supported by the Guangxi Natural Science Foundation Project (2023GXNSFAA026053); the Development, Popularization, and Application of Appropriate Medical and Health Technology in Guangxi (S2021090); and the National Natural Science Foundation of China (NSFC81360220, 81260214).

Data availability

All data generated or analysed during this study are included in this published article and its supplementary information files.

Declarations

Competing interests

The authors declare no competing interests.

Ethical approval

This research involved human participants. All data were obtained from routine clinical tests, and there was no clinical intervention for the participants in the study. This was a prospective study conducted in accordance with the Declaration of Helsinki and was approved by the Institutional Review Board of the First Affiliated Hospital of Guangxi Medical (Approval Number: 2024-E432-01).

Informed consent

All patients provided their written informed consent.