Qualitative and quantitative assessment of intratumoral fat using chemical-shift MRI for predicting histological grade of hepatocellular carcinoma

Abstract

Objectives

To evaluate intratumoral fat in hepatocellular carcinoma (HCC) using both qualitative and quantitative approaches based on routine chemical-shift magnetic resonance imaging (MRI), and to investigate its potential value in predicting histological grade.

Methods

This retrospective study included 282 patients with pathologically confirmed HCC between January 2015 and November 2025. Tumors were classified into low-grade and high-grade groups according to the Edmondson–Steiner grade. Intratumoral fat was assessed on in-phase and opposed-phase MRI images. For qualitative assessment, intratumoral fat pattern was categorized as none, heterogeneous, or homogeneous. For quantitative assessment, regions of interest were manually delineated on three consecutive slices showing the largest tumor area, and the mean fat fraction (FF) was calculated. Logistic regression analysis was performed to identify risk factors associated with high-grade HCC. Furthermore, models incorporating clinicoradiological factors were developed for the preoperative prediction of HCC histological grade.

Results

Homogeneous intratumoral fat was more frequently observed in low-grade tumors than in high-grade tumors, and FF was significantly higher in low-grade tumors than in high-grade tumors. Both homogeneous intratumoral fat (odds ratio [OR] = 0.230 [0.097–0.514], P = 0.001) and FF (OR = 0.861 [0.811–0.907], P < 0.001) were identified as independent predictors of high-grade HCC. When combined with other clinicoradiological factors, the FF-based model showed better performance than the intratumoral fat pattern–based model (area under the receiver operating characteristic curve: 0.792 vs. 0.744, P = 0.024).

Conclusion

Intratumoral fat assessed using chemical-shift imaging provides a simple and noninvasive imaging biomarker for predicting the histological grade of HCC. Both homogeneous intratumoral fat and higher FF were associated with a lower risk of high-grade HCC, and the model based on quantitative assessment outperformed that based on qualitative evaluation.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12880-026-02543-5.

Introduction

Hepatocellular carcinoma (HCC) constitutes the predominant form of primary liver cancer [1, 2]. It is associated with poor prognosis, with a reported relative 5-year survival rate of approximately 20% [3, 4]. The histological grade of HCC, reflecting the biological behavior of the tumor, is a well-recognized prognostic factor. Specifically, patients with high-grade tumors are reported to carry a higher risk of postoperative recurrence and poor survival [5, 6]. Therefore, prediction of histological grade may contribute to the pretreatment assessment of tumor aggressiveness.

Intratumoral fat can be present in some HCC cases, potentially as a result of transient hypoxia during tumor development [7]. A previous pathological study demonstrated that, in small HCC, fatty change on light microscopy was more frequently observed in well-differentiated lesions than in moderately or poorly differentiated ones, and its frequency decreased with increasing tumor diameter [8]. These findings indicate that intratumoral fat in HCC may have the potential to reflect tumor aggressiveness. However, whether this relationship can be consistently reflected on imaging remains unclear.

Magnetic resonance imaging (MRI) is widely utilized in the diagnosis and staging of HCC, while also enabling the noninvasive assessment of intratumoral fat. Chemical-shift imaging with in-phase (IP) and opposed-phase (OP) sequences, routinely included in liver MRI protocols, can depict intralesional fat as a drop in signal intensity (SI) on OP images relative to IP images [9]. Previous studies have qualitatively assessed intratumoral fat in HCC based on chemical-shift imaging, suggesting its presence may be associated with a lower risk of microvascular invasion and a more favorable prognosis [10–12]. However, no clear association between intratumoral fat and histological grade was identified in these studies. Moreover, qualitative assessment of intratumoral fat demonstrates only moderate interobserver agreement, which highlights the need for more objective quantitative evaluation methods. Some recent studies employed MRI proton density fat fraction (PDFF) mapping to quantify intralesional fat, and found PDFF values differed significantly among HCCs with different histological grades [13, 14]. Nevertheless, PDFF has not yet been included in routine scanning protocols, limiting its clinical use. Alternatively, fat fraction (FF) derived from IP and OP images provides an approach for quantifying hepatic steatosis, and its feasibility in detecting intratumoral fat in HCC has also been demonstrated [15, 16]. Despite this, FF calculated from IP and OP images may differ from the actual PDFF. Therefore, its utility for predicting HCC histological grade and its performance compared with conventional qualitative assessment have not been well established.

Hence, this study aimed to evaluate intratumoral fat in HCC using both qualitative and quantitative approaches based on IP and OP images and to explore its potential association with histological grade. Furthermore, we attempted to develop preliminary models integrating multiple clinical and imaging factors for preoperative prediction of histological grade.

Materials and methods

This retrospective study received approval from the institutional medical ethics committee (M20250016). Informed consent was waived based on the study design.

Study population

We searched the institutional pathology database to identify patients with pathologically confirmed HCC between January 2015 and November 2025. The inclusion criteria were: (1) age ≥ 18 years; (2) pathological confirmation of HCC. The exclusion criteria were: (1) absence of preoperative MRI or incomplete MRI sequences; (2) inadequate MRI image quality for analysis, as determined by consensus between the two radiologists who performed the image analysis; (3) maximum tumor diameter < 1 cm; (4) incomplete clinicopathological information; (5) complete intratumoral hemorrhage, precluding evaluation of intratumoral fat; (6) preoperative locoregional antitumor treatment of the target lesion. A flowchart illustrating the patient selection workflow is shown in Fig. 1. A total of 282 patients were ultimately included in the final analysis and were stratified into low- and high-grade groups according to tumor histological grade.

Fig. 1.

Flowchart of the patient selection process

Clinical and laboratory characteristics

Baseline clinical information and laboratory findings were collected from electronic medical records. Laboratory parameters included alpha-fetoprotein (AFP), viral hepatitis status, as well as basic liver function and routine blood tests. Inflammatory markers, including the aspartate aminotransferase (AST) to neutrophil ratio index (ANRI) and aspartate aminotransferase to lymphocyte ratio index (ALRI), were calculated as follows: ANRI = AST (U/L)/ neutrophil count (10⁹/L), ALRI = AST (U/L)/ lymphocyte count (10⁹/L) [17].

MRI acquisition

MRI examinations were performed using 1.5-T or 3.0-T MRI scanners from different manufacturers. One of the following liver MRI examinations was performed: non-contrast MRI, contrast-enhanced MRI with conventional extracellular contrast agents, or with hepatocyte-specific contrast agents. All examinations included at least T2-weighted imaging, T1-weighted imaging, and IP/OP imaging. Detailed scanner information and MRI parameters are provided in Table S1.

Image analysis

All MRI images were independently reviewed by two trained radiologists (with 5 and 7 years of experience, respectively) using a picture archiving and communication system (PACS; United Imaging Healthcare, China), blinded to the clinical and pathological information. The evaluation process of intratumoral fat is illustrated in Fig. 2. In patients with multiple tumors, one dominant lesion corresponding to the pathological findings was selected. For qualitative assessment, the presence of intratumoral fat was defined according to the 2018 Liver Imaging Reporting and Data System (LI-RADS) criteria as fat within a lesion that is entirely or partially in excess relative to the adjacent liver on IP and OP images [9]. Specifically, intratumoral fat was further classified into homogeneous and heterogeneous types. The homogeneous pattern was defined as a diffuse SI drop on OP images relative to IP images, without mosaic architecture or a nodule-in-nodule appearance [12]. Otherwise, the pattern was considered heterogeneous. All cases were independently evaluated by two readers, and disagreements were settled through discussion. The interobserver agreement between the two readers was subsequently evaluated. For quantitative assessment, FF based on IP and OP images was calculated using the following formula: FF= (SI_IP – SI_OP)/(2×SI_IP) [15, 18]. On OP images, the slice showing the largest tumor area was selected visually, and a region of interest was manually outlined along the tumor margin for SI_OP measurement. The measurements were repeated on the adjacent upper and lower slices, and the mean SI from three consecutive slices was calculated as the final value for analysis. To ensure consistent localization, the regions of interest drawn on the OP images were copied to the corresponding in-phase images, where SI_IP was measured using the same method [16]. All measurements were independently conducted by the radiologist with 5 years of experience, and a random subset of 40 patients was remeasured by a second radiologist to assess interobserver agreement.

Fig. 2.

Schematic illustration of qualitative and quantitative assessment of intratumoral fat. FF, fat fraction; IP, in-phase; OP, opposed-phase; SI, signal intensity

The following additional radiological features were also recorded in this study: (a) maximum tumor diameter, measured as the largest diameter on axial T2-weighted images. (b) tumor margin, classified as smooth or non-smooth; a non-smooth margin was defined as irregular contours with the presence of nodular or budding portions along the tumor boundary, whereas a smooth margin was defined as a round or oval tumor with a smooth contour [19, 20]; (c) intratumoral hemorrhage, defined as high SI within the tumor on fat-suppressed T1-weighted images [21]; (d) extratumoral features, including cirrhosis, macrovascular invasion, ascites, and enlarged lymph nodes, were recorded from the original radiology reports.

Pathological assessment

The histological grade of HCC was determined according to the Edmondson-Steiner grading system, based on pathology reports when available or re-evaluated by an experienced pathologist otherwise. Tumors with grades I, I–II and II were categorized as low grade, while those with grades II–III, III, III–IV and IV were categorized as high grade [22].

Statistical analysis

Statistical analyses were performed using SPSS 25 (IBM Corp., Armonk, NY, USA) and R 4.4.1 (R Foundation for Statistical Computing, Vienna, Austria). Continuous variables were summarized as mean ± standard deviation or median (interquartile range) and compared using the Student’s t-test or Mann–Whitney U test. Categorical variables were summarized as frequencies and percentages and compared using the chi-square test. Interobserver agreement for intratumoral fat was evaluated using Cohen’s kappa coefficient (κ) for qualitative assessment and intraclass correlation coefficients (ICC, two-way mixed, absolute agreement) for quantitative assessment. Associations between tumor size and intratumoral fat characteristics were assessed using Spearman’s rank correlation or non-parametric tests, where appropriate. Factors associated with high-grade tumors were explored using univariable and multivariable logistic regression analyses. Variables with P < 0.05 in univariable analysis were included in the multivariable analysis. Predictive models for HCC histological grade were constructed, with their performance evaluated using receiver operating characteristic (ROC) curves, calibration curves and decision curve analysis. Model performance was compared using the DeLong test. P value < 0.05 was considered statistically significant, with a Bonferroni-corrected threshold of P < 0.017 applied for pairwise comparisons.

Results

Baseline characteristics

Among the 282 patients included in the final analysis, 129 (45.74%) had low-grade tumors and 153 (54.26%) had high-grade tumors. The mean age of the study population was 61.50 ± 11.81 years, of whom 218 (77.30%) were male. The baseline clinical, laboratory, and imaging characteristics of patients in the low- and high-grade groups are summarized in Table 1. Intergroup comparisons revealed that patients in the high-grade group were slightly younger and had a lower body mass index than those in the low-grade group (both P < 0.05). In terms of laboratory findings, patients in the high-grade group showed higher lnAFP levels, along with higher proportions of HBsAg positivity and AST > 40 U/L. In addition, inflammatory markers, including ANRI and ALRI, were also higher in the high-grade group (all P < 0.05). Regarding imaging features, tumors in the high-grade group were larger in diameter. Moreover, the proportions of nonsmooth tumor margins, intratumoral hemorrhage, cirrhosis, macrovascular invasion, and enlarged lymph nodes were higher compared with the low-grade group (all P < 0.05). There were no significant differences in other baseline characteristics between the two groups.

Table 1.

Baseline characteristics of the study population

			Low-grade (n = 129)	High-grade (n = 153)	P value
Age, years			63.05 ± 10.51	60.18 ± 12.69	0.039
Gender					0.623
male			98(75.97%)	120(78.43%)
female			31(24.03%)	33(21.57%)
Body mass index, kg/m2			24.97 (22.84, 27.34)	24.22 (22.05, 26.23)	0.047
Laboratory findings
lnAFP, ng/ml			1.65 (0.97, 3.62)	3.95(1.76, 6.53)	< 0.001
HBsAg					0.046
Positive			71(55.04%)	102(66.67%)
Negative			58(44.96%)	51(33.33%)
HCVAb					0.199
Positive			6(4.65%)	13(8.50%)
Negative			123(95.35%)	140(91.50%)
Hemoglobin, g/L			143 (134, 153)	142 (131, 154)	0.780
Platelet, 109/L			175 (124, 209)	159 (123, 216)	0.583
Neurophil, 109/L			3.47 (2.73, 4.51)	3.2 (2.38, 4.48)	0.138
Lymphocyte, 109/L			1.46 (1.14, 1.82)	1.44 (1.05, 1.84)	0.583
Total bilirubin, µmol/L			15.2 (11.9, 18.8)	15.6 (12.5, 20.4)	0.204
Total protein, g/L			71.27 ± 6.12	72.02 ± 6.34	0.316
Albumin, g/L			42.5 (39.1, 44.8)	41.6 (39.3, 44.3)	0.261
ALT, U/L					0.098
≤ 50			111(86.05%)	120(78.43%)
> 50			18(13.95%)	33(21.57%)
AST, U/L					0.012
≤ 40			98(75.97%)	95(62.09%)
> 40			31(24.03%)	58(37.91%)
ANRI			8.42(5.73, 12.57)	10.43 (7.12, 18.12)	0.002
ALRI			19.01(14.37, 29.71)	26.32 (15.42, 41.82)	0.004
Imaging features
Tumor diameter, cm			3.5 (2.3, 5.6)	4.4 (2.7, 7.4)	0.011
Tumor margin					0.001
Smooth			73(56.59%)	55(35.95%)
Nonsmooth			56(43.41%)	98(64.05%)
Intratumoral hemorrhage					0.044
Absent			97(75.19%)	98(64.05%)
Present			32(24.81%)	55(25.95%)
Cirrhosis					0.027
Absent			81(62.79%)	76(49.67%)
Present			48(37.21%)	77(50.33%)
Macrovascular invasion					0.007
Absent			125(96.90%)	135(88.24%)
Present			4(3.10%)	18(11.76%)
Ascites					0.174
Absent			113(87.60%)	125(81.70%)
Present			16(12.40%)	28(18.30%)
Enlarged lymph nodes					0.023
Absent			113(87.60%)	118(77.12%)
Present			16(12.40%)	35(22.88%)

AFP, alpha-fetoprotein; ALRI, aspartate aminotransferase to lymphocyte ratio index; ALT, alanine aminotransferase; ANRI, aspartate aminotransferase to neutrophil ratio index; AST, aspartate aminotransferase; HBsAg, hepatitis B virus surface antigen; HCVAb, hepatitis C virus antibodies

Qualitative and quantitative assessment of intratumoral fat on MRI

Table 2 summarizes the qualitative and quantitative MRI-assessed intratumoral fat. For qualitative assessment, a higher proportion of overall intratumoral fat was observed in the low-grade group compared with the high-grade group (57.36% vs. 41.83%; P = 0.009). Intratumoral fat patterns were further compared between the two groups and categorized into three types: no fat, heterogeneous fat, and homogeneous fat. A significant difference in the distribution of intratumoral fat patterns was also observed between the two groups (P < 0.001; Table 2). Pairwise comparisons showed that tumors with homogeneous intratumoral fat had a higher proportion of low-grade HCC than those with no fat or heterogeneous intratumoral fat (both P < 0.017; Fig. 3A). No significant difference was observed between tumors with no fat and those with heterogeneous intratumoral fat (P = 0.319; Fig. 3A). For quantitative assessment, the FF was significantly higher in the low-grade group than in the high-grade group (7.94 [4.82, 12.81] vs. 4.64 [2.04, 7.26]; P < 0.001; Fig. 3B).

Table 2.

Qualitative and quantitative assessment of intratumoral fat on MRI

		Low-grade (n = 129)	High-grade (n = 153)	P value
Qualitative assessment
Overall intratumoral fat				0.009
Absent		55(42.64%)	89(58.17%)
Present		74(57.36%)	64(41.83%)
Intratumoral fat pattern				< 0.001
No		55(42.64%)	89(58.17%)
Heterogeneous		42(32.56%)	52(33.99%)
Homogeneous		32(24.81%)	12(7.84%)
Quantitative assessment
FF, %		7.94 (4.82, 12.81)	4.64 (2.04, 7.26)	< 0.001

FF, fat fraction

Fig. 3.

Intratumoral fat patterns and FF in low- and high-grade tumors. (A) Distribution of intratumoral fat patterns in low- and high-grade tumors (Bonferroni-corrected significance level: P < 0.017). (B) Comparison of FF between low- and high-grade tumors. FF, fat fraction

A moderate inter-observer agreement was observed for intratumoral fat pattern (Cohen’s κ = 0.45; 95% CI = 0.37–0.53). In a randomly selected subset of 40 patients, quantitative FF measurements demonstrated excellent inter-observer agreement (ICC = 0.94; 95% CI = 0.88–0.97). Figure 4 illustrates representative cases of HCC lesions with different intratumoral fat characteristics.

Fig. 4.

Representative cases of HCC lesions with different intratumoral fat characteristics. (A-D) Patient (1) T2-weighted imaging shows a lesion measuring approximately 2.4 cm in the right hepatic lobe (A). IP and OP images demonstrate homogeneous intratumoral fat, with a mean FF of approximately 20.8% (B, C). H&E staining (×20) shows an Edmondson–Steiner grade of 1–2 (D). (E-H) Patient (2) T2-weighted imaging shows a lesion measuring approximately 5.6 cm in the left hepatic lobe (E). IP and OP images demonstrate no intratumoral fat, with a mean FF of approximately 4.5% (F, G). H&E staining (×20) shows an Edmondson–Steiner grade of 3–4 (H). (I-L) Patient (3) T2-weighted imaging shows a lesion measuring approximately 2.9 cm in the left hepatic lobe (I). IP and OP images demonstrate heterogeneous intratumoral fat, with a mean FF of approximately 8.1% (J, K). H&E staining (×20) shows an Edmondson–Steiner grade of 1–2 (L). IP, in-phase; OP, opposed-phase; FF, fat fraction

Association between tumor size and intratumoral fat

Figure 5 illustrates the association between tumor diameter and intratumoral fat characteristics. Significant differences in tumor diameter were observed across the three qualitative intratumoral fat pattern groups (no fat: 4.05 [2.6–6.55] cm; heterogeneous fat: 4.9 [3.08–8.55] cm; homogeneous fat: 2.7 [1.9–4.7] cm; P < 0.001). Post hoc analyses demonstrated significant differences in tumor diameters among all pairwise comparisons, with the smallest diameters observed in tumors with homogeneous intratumoral fat and the largest in tumors with heterogeneous fat (all P < 0.017; Fig. 5A). However, no significant correlation was observed between tumor diameter and FF (P = 0.836; Fig. 5B).

Fig. 5.

Association between tumor diameter and intratumoral fat characteristics. (A) Qualitative intratumoral fat pattern. (B) Quantitative FF measurement. FF, fat fraction; rₛ, Spearman correlation coefficient

Independent predictors of HCC histological grade

The results of univariable and multivariable logistic regression analyses are presented in Table S2 and Table 3. Based on variables identified as significant in univariable analyses, two separate multivariable analyses were performed to further identify independent predictors of high-grade HCC. Multivariable analysis 1 incorporated qualitative intratumoral fat patterns, whereas Multivariable analysis 2 incorporated quantitative FF. In multivariable analysis 1, a homogeneous intratumoral fat pattern (odds ratio [OR]: 0.230, 95% confidence interval [CI]: 0.097–0.514; P = 0.001), lnAFP ( OR: 1.231, 95% CI: 1.109–1.374; P < 0.001), and a nonsmooth tumor margin (OR: 1.870, 95% CI: 1.066–3.310; P = 0.030) remained independent predictors of high-grade HCC. In multivariable analysis 2, FF (OR: 0.861, 95% CI: 0.811–0.907; P < 0.001), lnAFP (OR: 1.254, 95% CI: 1.123–1.411; P < 0.001) and a nonsmooth tumor margin (OR: 1.833, 95% CI: 1.018–3.327; P = 0.044) remained independent predictors of high-grade HCC (Table 3).

Table 3.

Multivariable logistic regression analyses for high-grade HCC

		Multivariable analysis 1		Multivariable analysis 2
		Odds ratio (95% CI)	p value	Odds ratio (95% CI)	p value
Intratumoral fat pattern				-	-
No		Reference		-	-
Heterogeneous		0.774[0.426, 1.404]	0.398	-	-
Homogeneous		0.230[0.097, 0.514]	0.001	-	-
FF, %		-	-	0.861[0.811, 0.907]	< 0.001
Age, years		0.995[0.970, 1.021]	0.696	0.998[0.972, 1.025]	0.887
ln AFP, ng/ml		1.231[1.109, 1.374]	< 0.001	1.254[1.123, 1.411]	< 0.001
HBsAg, positive		1.483[0.819, 2.696]	0.193	1.546[0.830, 2.894]	0.170
AST, > 40 U/L		0.830[0.391, 1.719]	0.620	0.886[0.397, 1.929]	0.764
ANRI		1.018[0.992, 1.054]	0.245	1.016[0.989, 1.054]	0.327
Tumor margin, nonsmooth		1.870[1.066, 3.310]	0.030	1.833[1.018, 3.327]	0.044
Intratumoral hemorrhage, present		1.288[0.671, 2.481]	0.447	1.531[0.770, 3.079]	0.226
Cirrhosis, present		1.636[0.916, 2.941]	0.097	1.496[0.817, 2.756]	0.193
Macrovascular invasion, present		1.705[0.543, 6.538]	0.390	1.835[0.541, 7.702]	0.361
Enlarged lymph nodes, present		1.211[0.572, 2.615]	0.619	1.096[0.500, 2.445]	0.819

Note: For continuous variables, odds ratios are expressed per one unit increase in the corresponding variable

AFP, alpha-fetoprotein; ANRI, aspartate aminotransferase to neutrophil ratio index; AST, aspartate aminotransferase; CI, confidence interval; FF, fat fraction; HBsAg, hepatitis B virus surface antigen

Development and evaluation of prediction models

Based on the identified independent predictors, two models were constructed respectively. Model 1 included qualitatively assessed intratumoral fat patterns, lnAFP, and tumor margin, whereas Model 2 included quantitatively measured FF, lnAFP, and tumor margin. The predictive performance of the two models is shown in Table 4; Fig. 6. ROC analysis demonstrated that the area under the curve (AUC) of Model 2 was significantly higher than that of Model 1(0.792 vs. 0.744, P = 0.024) (Fig. 6A). Using the optimal cut-off values determined by the Youden index, Model 2 achieved a sensitivity of 83.66% with a corresponding specificity of 61.24%, while Model 1 demonstrated a more balanced performance, with a sensitivity of 72.55% and a specificity of 69.77% (Table 4).

Table 4.

Predictive performances of the models

Models	AUC	Cut-off	Sensitivity (%)	Specificity (%)
Model 1*	0.744	0.52	72.55%	69.77%
Model 2†	0.792	0.44	83.66%	61.24%

^*Model 1 included qualitatively assessed intratumoral fat patterns, lnAFP, and tumor margin

^†Model 2 included quantitatively measured FF, lnAFP, and tumor margin

AFP, alpha-fetoprotein; AUC, area under the receiver operating characteristic curve; FF, fat fraction

Fig. 6.

Performance comparison between the two models. (A) ROC curves. (B) Decision curve analysis illustrating the net benefit of the models. (C) Calibration curves showing the agreement between predicted and observed probabilities. Model 1 included qualitatively assessed intratumoral fat patterns, lnAFP, and tumor margin; Model 2 included quantitatively measured FF, lnAFP, and tumor margin. AFP, alpha-fetoprotein; FF, fat fraction; ROC, receiver operating characteristic

Decision curve analysis showed that both Model 1 and Model 2 provided higher net benefit compared with the treat-all and treat-none across a threshold probability range of approximately 0.2–0.8, with Model 2 yielding a higher net benefit than Model 1(Fig. 6B). Calibration curves indicated satisfactory consistency between the predicted and actual probabilities in both models (Fig. 6C).

Discussion

Preoperative noninvasive prediction of histological grade is crucial for individualized HCC management. However, previous MRI-based histological grading models mainly relied on conventional or hepatocyte-specific contrast-enhanced MRI, which may not be feasible for all patients [21, 23–25]. Given that intratumoral fat can be assessed on non-contrast MRI, exploring its potential utility in predicting the biological behavior of HCC is of significant clinical importance.

The association between imaging-assessed intratumoral fat and the histological grade of HCC remains controversial. In this work, we assessed intratumoral fat in HCC using both qualitative and quantitative methods based on IP and OP images, and investigated its relationship with histological grade. Our findings showed that both homogeneous intratumoral fat and FF were recognized as independent predictors of the histological grade of HCC. Furthermore, predictive models integrating qualitative or quantitative fat assessment with clinicoradiological factors were developed, and the model based on FF demonstrated better performance than that based on qualitative assessment.

Visually assessed intratumoral fat has been reported in about 18%–40% of HCCs on IP and OP images [11, 12, 26]. In our cohort, it was observed in approximately 48% of cases, which is higher than previously reported. This difference may be attributable to variations in the study population and assessment standards, as well as the subjectivity of intratumoral fat evaluation. It has been hypothesized that the formation of intratumoral fat is related to dynamic changes in tumor perfusion. During the early development of HCC, transient ischemia may occur as the tumor blood supply transitions from portal venous to arterial perfusion, facilitating the accumulation of intratumoral fat. In the late stage of HCC, arterial blood flow may decrease again, potentially contributing to fatty changes in advanced stage [8, 26, 27]. Our results showed that homogeneous intratumoral fat was more commonly observed in low-grade tumors than in high-grade tumors, and FF was also significantly higher in low-grade tumors, consistent with previous pathological findings [8]. One possible explanation is that lipid metabolic reprogramming may occur during the development of HCC. Some studies have suggested that the expression of enzymes involved in fatty acid synthesis may be enhanced in well-differentiated HCC, and that fatty acid synthesis and uptake are most active in well-differentiated HCC but decrease with tumor dedifferentiation [28, 29].

Previous studies investigating intratumoral fat in HCC have mainly relied on qualitative assessment on IP and OP images, which has shown clinical value but remains subjective. Jiang et al. [12] reported that homogeneous intratumoral fat was associated with longer progression-free survival and overall survival, but the interobserver agreement was only moderate (Fleiss κ, 0.38–0.56). A similar level of agreement was observed in our study for qualitative intratumoral fat assessment (Cohen’s κ = 0.45). Our study further incorporated quantitative FF measurement from IP/OP images, which demonstrated excellent interobserver agreement and provided objective information on intratumoral fat content beyond visual assessment. When combined with other clinicoradiological features, the FF-based model achieved better diagnostic performance than the qualitative model (AUC: 0.792 vs. 0.744, P = 0.024) for predicting HCC histological grade, indicating the additional diagnostic value of quantitative fat assessment. This may be partly because qualitative assessment mainly captures fat heterogeneity, whereas quantitative FF measurement more directly represents fat content.

PDFF is a reliable quantitative biomarker for hepatic fat assessment [30]. Several studies have suggested the value of PDFF-based intratumoral fat quantification for differentiating the histological grade of HCC [13, 14]. However, these studies were limited by relatively small sample sizes, imbalanced sample distributions, or insufficient diagnostic performance of PDFF as a single predictor. In addition, PDFF mapping is not yet widely adopted in clinical practice. Compared with these studies, the added value of our study lies in the use of FF derived from routine IP and OP images. This allowed our analysis to be performed in a relatively larger cohort. Furthermore, by integrating FF with two easily available clinicoradiological features, lnAFP and tumor margin, our combined model showed better performance than a previously reported model based solely on PDFF value (AUC: 0.792 vs. 0.683) [14].

MRI-based radiomics and deep learning approaches have also been applied to predict HCC histological grade, with favorable performance reported in validation sets, where AUCs are generally around 0.80 [22, 23, 25, 31]. Nevertheless, many of these models rely on multiphase contrast-enhanced or hepatobiliary phase MRI and have limited clinical interpretability. By comparison, our approach was less dependent on contrast administration or contrast-agent type. In addition, our model did not require complex image segmentation, feature extraction, or algorithm development, making it easier to implement in routine clinical practice [32]. The variables included in our model all have clear clinical relevance and are readily interpretable. Although direct cross-study comparison is inappropriate, our model achieved an AUC of 0.792, suggesting that this simpler workflow may have clinical value for routine preoperative histological grade prediction.

Notably, the FF in our study was calculated from the SI difference between IP and OP images. Although it may partially reflect intratumoral fat-related signal loss, it does not represent PDFF obtained from confounder-corrected multi-echo chemical-shift encoded MRI. Conventional dual-echo IP/OP imaging is vulnerable to several technical sources of bias. It mainly reflects the signal contribution of the dominant CH₂ fat peak, while other minor fat peaks are not accounted for, potentially leading to underestimation of FF [33]. In addition, uncorrected T2* decay in the dual-echo method may also contribute to FF underestimation, particularly in HCC lesions with iron deposition [15, 16]. Conversely, T1 bias may introduce overestimation of FF, the extent of which depends on flip angle and repetition time [15, 34–36]. In HCC lesions, these biases may be more complex because of heterogeneous tumor composition [16]. Despite these technical limitations compared with PDFF, IP/OP-derived FF may still have clinical utility, especially when PDFF sequences are not available. Its main advantage is that it can be easily and rapidly derived from standard liver MRI protocols, making it practical for incorporation into routine radiological workflows. Radiologists can perform FF measurements directly on the radiology workstation and integrate the results with other clinical and radiological information to support a more comprehensive preoperative grade assessment.

Our study has several limitations. First, this was a single-center retrospective study, and selection bias could not be avoided. As an exploratory study, our findings require validation in independent cohorts. Second, sufficient clinical outcome data were not available for analysis, precluding further exploration of the associations between intratumoral fat characteristics and patient outcomes. Therefore, our findings are limited to the prediction of histological grade and should not be extended to clinical outcomes. We will collect follow-up data to analyze clinical outcomes in future studies. Third, MRI examinations were acquired using scanners from different vendors and field strengths, although the majority were acquired at 3.0 T. However, no phantom calibration or harmonization procedure was performed, and therefore potential scanner-related variability in FF measurement could not be completely eliminated. The excellent inter-observer ICC of FF supports measurement reproducibility between readers but cannot compensate for scanner-related variability. To partially address this limitation, we performed a sensitivity analysis restricted to the scanner used for the largest number of patients, and FF remained significantly higher in low-grade HCC than in high-grade HCC in this subgroup. Future studies with standardized acquisition and harmonization methods are needed to validate the generalizability of FF across scanners. Fourth, IP/OP-derived FF is not equivalent to PDFF and may be affected by technical biases. Fifth, patients with multiple tumors were not excluded. Although the dominant lesion was selected for imaging analysis, laboratory parameters may reflect the overall tumor burden. Finally, contrast-enhanced imaging features were not included, and future studies may integrate intratumoral fat assessment with multimodal imaging to further improve predictive accuracy [37].

Conclusions

In conclusion, intratumoral fat assessed using chemical-shift imaging may serve as a simple and noninvasive imaging biomarker for predicting the histological grade of HCC. Both homogeneous intratumoral fat on qualitative assessment and higher FF on quantitative assessment are associated with a lower risk of high-grade tumors. Combined models integrating intratumoral fat characteristics with clinicoradiological variables may support preoperative prediction of tumor grade in HCC, with quantitative fat assessment demonstrating better predictive performance than qualitative evaluation.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

AFP

Alpha-fetoprotein

ALRI

Aspartate aminotransferase to lymphocyte ratio index

ANRI

Aspartate aminotransferase to neutrophil ratio index

AST

Aspartate aminotransferase

AUC

Area under the receiver operating characteristic curve

CI

Confidence interval

FF

Fat fraction

ICC

Intraclass correlation coefficient

HCC

Hepatocellular carcinoma

IP

In-phase

LI-RADS

Liver Imaging Reporting and Data System

MRI

Magnetic resonance imaging

OP

Opposed-phase

OR

Odds ratio

PDFF

Proton density fat fraction

ROC

Receiver operating characteristic

SI

Signal intensity

Author contributions

XD and QW contributed to the conception, data curation, formal analysis and drafting the work. YH, JY and JS contributed to pathological analysis. LF, JX, CL and DJ assisted with methodology and investigation. NL supervised the study and revised the work. All authors read and approved the final manuscript.

Funding

This work was supported by the Beijing Medical Award Foundation [grant number YXJL-2025-0483-0215].

Data availability

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This retrospective study was approved by the Ethics Committee of Peking University Third Hospital (M20250016), and the informed consent was waived based on the study design. The study was conducted in accordance with the Declaration of Helsinki.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.