Diagnostic value of CT-quantified abdominal fat parameters for metabolic dysfunction-associated fatty liver disease

Abstract

Background

To analyze the diagnostic value of CT-derived abdominal fat indices in predicting MAFLD and establish an integrated diagnostic system.

Methods

A retrospective study on 174 patients with a previous history of abdominal CT examination in our institute between January 2022 and June 2024 was conducted, comprising 120 diagnosed with MAFLD, and 54 controls without it. CT-derived measures including SAT, visceral adipose tissue volume(VAT), total liver volume, and average hepatic. The average value and CT attenuation value of the left and right lobe of liver and the VAT/SAT were compared using Mann-Whitney U test. were compared between both groups. Then, we used Spearman’s rank correlation analysis to study whether there existed correlations between CT metrics and metabolic indicators. Logistic regression and receiver operating characteristic(ROC)curve analysis were utilized to evaluate the diagnostic accuracy of individual parameters.Calibration of a composite prediction model was evaluated by using calibration plots.

Results

In MAFLD groups, SAT, VAT, liver volume, and the VAT / SAT ratio were higher than those of the control group, while the mean hepatic CT attenuation value, as well as the mean attenuation value of the left lobe and right lobe, were all lower than those of the control group (all P < 0.05). The mean hepatic CT attenuation correlated negatively with BMI (r = -0.607) and TG (r = -0.418), while the sign was positive for VAT. In univariate logistic regression, we found that the mean hepatic CT attenuation is the best diagnostic indicator of MAFLD, with an AUC of 0.923, then were left and right hepatic attenuation (AUC = 0.886), liver volume (AUC = 0.814), and VAT’s mean CT attenuation (AUC = 0.796). BMI (r = 0.554) and UA (r = 0.362) were also considered. Combining SAT, VAT, and mean hepatic CT attenuation together in one model resulted in an AUC of 0.928 (sensitivity: 91.7%, specificity: 85.2%). The calibration curves demonstrated an acceptable fit of the model. When multiple variables were combined, the sensitivity improved from 85.8% to 91.7% and specificity changed from 87.0% to 85.2%, yielding an AUC of 0.928 (ΔAUC = 0.005 compared with mean hepatic CT attenuation alone, P = 0.62).

Conclusion

CT-measured abdominal fat may represent a reliable and noninvasive approach for the diagnosis of MAFLD.The greatest individual diagnostic utility was exhibited by mean liver CT attenuation alone; however, with multiple variables merged into a combined model, the sensitivity and specificity were also further increased, which could be an effective method to image MAFLD in clinical diagnosis.

Introduction

First proposed the name of metabolic dysfunction-associated fatty liver disease(MAFLD) in 2020, and Fan disseminated it to the medical community across China [1, 2]. The new name for fatty liver is widely recognized.MAFLD represents a change from an exclusive diagnosis of NAFLD to one that is more positive, metabolism focused one which reflects the mechanism of the disease accurately [3]. The prevalence of NAFLD has been estimated to be approximately 25% worldwide [4], and in China, it increased from 23.8% to 32.9% [5]. From 1999 to 2018, the prevalence of fatty liver as the leading chronic liver condition in the nation increased by 9%. Not only does MAFLD cause liver-related complications such as fibrosis and cirrhosis, but it simultaneously increases cardiovascular risk through insulin resistance and systemic inflammation [6]^, which is considered one of the most important public health problems. To unify the clinical approach, Chinese guidelines for NAFLD and MAFLD have been published [7, 8].

CAP of ultrasound elastography is widely used as a noninvasive diagnostic tool but the accuracy in obese patients or operator dependency [9] limits its applicability. On the other hand, unenhanced CT is readily available and highly standardized, with a single scan providing both hepatic density values and quantification of abdominal fat depots [10–12]. This research aimed to evaluate the diagnostic value of several CT-quantitative abdominal fat parameters in MAFLD diagnosis, as well as to establish and validate a combined logistic regression model. In contrast, most previous studies investigated a single CT parameter only, while few attempts were made to integrate hepatic density indicators with abdominal fat distribution indicators in a comprehensive diagnostic model of MAFLD. The goal of developing a new imaging-based method for noninvasive evaluation of MAFLD will be achieved.

Beyond CAP, a broader spectrum of noninvasive imaging modalities is currently used for hepatic steatosis assessment, each with distinct advantages and limitations. Conventional B-mode ultrasound (US) remains the first-line screening tool because of its accessibility and cost, but its qualitative grading suffers from operator dependency and limited sensitivity in mild steatosis [13].

Quantitative US-based approaches such as the hepatorenal index, backscatter coefficient, and attenuation imaging have improved reproducibility and correlate well with MR proton density fat fraction in recent prospective work [14], yet these remain dependent on vendor-specific software and operator expertise. MRI-based proton density fat fraction (MRI-PDFF) is considered the most accurate noninvasive reference standard, providing voxel-wise quantification with excellent reproducibility, but its cost, scanning time, and limited availability restrict large-scale screening use [15]. Against this background, the opportunistic use of routine unenhanced abdominal CT is especially appealing: CT scans performed for unrelated clinical indications already contain rich information on hepatic attenuation, visceral and subcutaneous adipose tissue, and skeletal muscle, which can be retrieved retrospectively without additional radiation exposure or cost to the patient.

This opportunistic strategy converts every clinically indicated abdominal CT into a potential metabolic phenotyping examination, supports population-level screening for MAFLD in patients who would otherwise not undergo dedicated steatosis imaging, and is increasingly endorsed for cardiometabolic risk stratification.

Materials and methods

Study population

We retrospectively reviewed data from patients who underwent abdominal CT at the Affiliated Hospital of Jiangsu University between January 2022 and June 2024. Patients were included if they (1) had undergone a standardized unenhanced abdominal CT and (2) had complete clinical and laboratory records. Patients were excluded if they (1) had chronic heavy alcohol use (weekly ethanol intake ≥ 210 g for men or ≥ 140 g for women) ; (2) had concurrent viral hepatitis, autoimmune liver disease, drug-induced liver injury, or other hepatic diseases; (3) had diffuse hepatic lesions or hepatic tumors; or (4) had CT images of insufficient quality for reliable measurement.

MAFLD was diagnosed per the 2020 international expert consensus, imaging evidence of hepatic steatosis (ultrasonography or CT indicating fatty liver) plus at least one of the following: (1) overweight or obesity (BMI ≥ 23 kg/m², Asian criteria); (2) type 2 diabetes mellitus; or (3) evidence of metabolic dysregulation. In total, 174 patients were enrolled, including 120 in the MAFLD group and 54 in the non-MAFLD control group. The study was approved by the institutional ethics committee. The enrollment window is identical in the abstract and the methods (January 2022 to June 2024); we have re-verified this throughout the manuscript. To address the age imbalance between groups, a sensitivity analysis was additionally performed in an age-stratified subset (40 to 60 years), in which the principal between-group comparisons of CT parameters and the discriminative performance of mean hepatic CT attenuation remained materially unchanged, suggesting that age-related differences were unlikely to fully explain the observed associations.

CT examination protocol and parameter measurement

Unenhanced abdominal CT was performed on a multidetector-row scanner (tube voltage 120 kV, automatic tube current modulation, slice thickness 5 mm, reconstruction thickness 1.25 mm).

All images were transferred to a postprocessing workstation for quantitative analysis. (1) Mean hepatic CT attenuation: regions of interest (ROIs, area ≥ 1 cm²) were placed in the hepatic parenchyma, avoiding vessels, bile ducts, and focal lesions, with one ROI in the left lobe and one in the right lobe; CT attenuation values (HU) were recorded and averaged. (2) SAT and VAT: at the L3 vertebral level, adipose tissue was quantified by semi-automated segmentation using a fat-density threshold of − 190 to − 30 HU. (3) Liver volume: total hepatic volume (cm³) was calculated via semi-automated segmentation. (4) The VAT/SAT ratio was computed from the above measurements.

All examinations were acquired on a single CT platform (Siemens SOMATOM Definition AS 128-slice; Siemens Healthineers, Erlangen, Germany) operated under a standardized abdominal protocol, so that inter-vendor variability did not contribute to the measurements reported here. Three circular ROIs were placed in each of the right (segments VI to VIII) and left (segments II to III) hepatic lobes at the level of the porta hepatis on the same axial slice, with each ROI exceeding 1 cm² (typically 1.0 to 1.5 cm²) and positioned at least 1 cm away from vessels, bile ducts, capsule, and focal lesions; the six values per patient were averaged to derive the mean hepatic CT attenuation, and lobar means were calculated separately. For livers showing diffuse heterogeneous attenuation, ROIs were redistributed to sample at least two non-contiguous regions per lobe at the same level, and an additional adjacent slice was sampled when local heterogeneity exceeded visual estimation; areas of focal fatty sparing, focal fatty infiltration, or perivascular artefact were systematically excluded. The hepatic-to-splenic attenuation ratio and the liver minus spleen HU difference (ΔHU) were additionally recorded for every patient using a single splenic ROI of comparable size placed in homogeneous splenic parenchyma at the same axial level. SAT and VAT were quantified at the L3 vertebral midbody on a single transverse slice using semi-automated thresholding (− 190 to − 30 HU) with manual correction of the abdominal wall boundary, and total liver volume was generated by semi-automated organ segmentation followed by slice-by-slice manual refinement. All measurements were performed independently by two radiologists with 8 and 12 years of abdominal imaging experience who were blinded to clinical and laboratory data; inter-observer agreement was quantified with the intraclass correlation coefficient (ICC, two-way random, absolute agreement), which exceeded 0.90 for hepatic attenuation and liver volume and exceeded 0.85 for SAT and VAT; the mean of the two readings was used for analysis. With respect to radiation, all scans had been clinically indicated for reasons unrelated to the present study; no additional acquisition was performed for research purposes, so no incremental radiation was delivered, and the recorded dose-length product fell within institutional and national reference levels for unenhanced abdominal CT.

Laboratory parameters

Fasting venous blood samples drawn within one week of the CT examination were analyzed for total cholesterol (TC), triglycerides (TG), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), uric acid (UA), alanine aminotransferase (ALT), aspartate aminotransferase (AST), gamma-glutamyl transferase (GGT), and fasting blood glucose (FBG).

Statistical analysis

All analyses were performed in R (version 4.3.0). Non-normally distributed continuous variables were reported as median (interquartile range) [M (Q1, Q3)] and compared with the Mann–Whitney U test. Categorical variables were expressed as counts (percentages) and compared with the chi-squared test. Spearman’s rank correlation was used to evaluate associations between CT parameters and metabolic indices. Univariate logistic regression was performed for each CT parameter with MAFLD status as the outcome; ROC curves were plotted and optimal cutoff values identified by maximizing the Youden index. A combined logistic regression model was built, and its performance was assessed with calibration curves. Agreement between CT attenuation values of the left and right hepatic lobes was evaluated by the Bland–Altman method.

A two-sided P < 0.05 was considered statistically significant. In addition to univariate analysis, a multivariable logistic regression model was fitted, including mean hepatic CT attenuation, VAT, SAT, liver volume, and age as candidate covariates, with age forced in to address the baseline imbalance; variables retained in the model are reported as adjusted odds ratios with 95% confidence intervals, and only these are interpreted as independent predictors of MAFLD.

Univariate results are presented as single-predictor associations rather than as independent risk factors. The discriminative performance of the combined model was directly compared with that of mean hepatic CT attenuation alone using the DeLong test for two correlated ROC curves, and Net Reclassification Improvement and Integrated Discrimination Improvement were computed as supportive metrics. Inter-observer agreement of all CT measurements was quantified using the intraclass correlation coefficient (ICC, two-way random, absolute agreement) with 95% confidence intervals. Because no external cohort was available, internal validation of the combined model was performed using 1000 bootstrap resamples, with optimism-corrected AUC and calibration slope reported.

Results

Baseline characteristics

A total of 174 patients were included. Baseline characteristics are summarized in Table 1.

Table 1.

Baseline characteristics of the two groups

Parameter	MAFLD group (n = 120)	Non-MAFLD group (n = 54)	Statistic	P value
Age (years)	48.2 ± 14.2	54.5 ± 15.9	Z = − 2.548	0.011
Sex (male/female)	70/50	34/20	χ² = 0.168	0.682
BMI (kg/m²)†	27.74 ± 3.84	24.39 ± 3.12	Z = − 4.650	< 0.001
TC (mmol/L)	5.02 (4.27, 6.41)	3.89 (3.35, 4.48)	Z = − 3.882	< 0.001
TG (mmol/L)	2.38 (1.62, 4.09)	1.07 (0.88, 1.45)	Z = − 5.419	< 0.001
HDL-C (mmol/L)	1.15 (0.96, 1.33)	1.20 (0.90, 1.47)	Z = − 0.693	0.489
LDL-C (mmol/L)	3.15 (2.58, 3.83)	2.29 (1.91, 3.04)	Z = − 3.862	< 0.001
UA (µmol/L)	348.0 (283.0, 419.0)	298.0 (220.3, 361.3)	Z = − 3.253	0.001
ALT (U/L)	30.0 (19.5, 52.0)	22.5 (15.0, 32.0)	Z = − 2.553	0.011
AST (U/L)	29.0 (22.0, 43.5)	27.5 (22.0, 43.8)	Z = − 0.014	0.989
GGT (U/L)	45.0 (29.0, 81.0)	30.5 (17.0, 59.0)	Z = − 2.740	0.006
FBG (mmol/L)	6.47 (5.45, 8.39)	5.76 (4.97, 6.68)	Z = − 2.910	0.004

Non-normally distributed continuous variables are reported as M (Q1, Q3) and compared with the Mann–Whitney U test; categorical variables were compared with the chi-squared test

^†BMI data were available for 100 patients in the MAFLD group and 48 in the non-MAFLD group

MAFLD patients were younger (P = 0.011) and had a higher BMI (P < 0.001) than controls. TG, TC, LDL-C, UA, ALT, GGT, and FBG were all elevated in the MAFLD group (all P < 0.05), whereas sex distribution, HDL-C, and AST did not differ between groups (all P > 0.05).

Comparison of CT-quantified parameters

CT parameters for both groups are listed in Table 2. SAT, VAT, liver volume, and the VAT/SAT ratio were all higher in the MAFLD group, while mean hepatic CT attenuation and the mean CT attenuation of the left and right lobes were lower (all P < 0.05). The largest between-group difference was seen for mean hepatic CT attenuation (P < 0.001). Box plots are presented in Fig. 1.

Table 2.

CT-quantified parameters in the two groups [M (Q1, Q3)]

CT Parameter	MAFLD group (n = 120)	Non-MAFLD group (n = 54)	Z value	P value
SAT (cm²)	88.0 (65.8, 119.2)	60.0 (53.2, 84.2)	−4.379	< 0.001
VAT (cm³)	112.0 (82.0, 135.2)	68.0 (52.0, 91.8)	−5.427	< 0.001
Liver volume (cm³)	1707.5 (1432.2, 2165.5)	1253.0 (1097.5, 1421.2)	−6.118	< 0.001
Mean hepatic CT attenuation (HU)	41.2 (33.4, 47.0)	52.5 (50.0, 57.5)	−7.891	< 0.001
Mean CT attenuation of L and R lobes (HU)	43.8 (34.5, 51.5)	57.5 (54.5, 63.0)	−7.234	< 0.001
VAT/SAT ratio	1.2 (0.8, 1.7)	1.0 (0.7, 1.4)	−2.117	0.034

Mann–Whitney U test

SAT  Subcutaneous adipose tissue area at the L3 level, VAT  Visceral adipose tissue volume at the L3 level

Fig. 1.

Box plots comparing CT-quantified parameters between MAFLD and non-MAFLD groups. A, Subcutaneous adipose tissue area; B, Visceral adipose tissue volume; C, Liver volume; D, Mean hepatic CT attenuation; E, Mean CT attenuation of the left and right hepatic lobes; F, VAT/SAT ratio

Correlations between CT parameters and metabolic indices

Spearman analysis (Fig. 2) showed that mean hepatic CT attenuation was negatively correlated with BMI (r = − 0.607, P < 0.001), TG (r = − 0.418, P < 0.001), ALT (r = − 0.381, P < 0.001), GGT (r = − 0.342, P < 0.001), and UA (r = − 0.298, P < 0.001). VAT was positively correlated with BMI (r = 0.554, P < 0.001), UA (r = 0.362, P < 0.001), and TG (r = 0.307, P < 0.001), while SAT correlated positively with BMI (r = 0.523, P < 0.001). A positive correlation between hepatic CT attenuation and HDL-C (r = 0.276, P < 0.001) was also observed, indicating that reduced hepatic CT density in fatty liver is linked to disordered lipid metabolism.

Fig. 2.

Spearman correlation heatmap. Red: positive correlation; blue: negative correlation. *P < 0.05, **P < 0.01, **P < 0.001

Diagnostic performance of individual CT parameters

Univariate logistic regression (Fig. 3) identified all CT parameters as single-predictor variables significantly associated with MAFLD status (all P < 0.05). Each 1-HU increase in mean hepatic CT attenuation was associated with a 28.4% decrease in MAFLD risk (OR = 0.716, 95% CI: 0.638–0.803, P < 0.001), and each 1-cm³ increase in VAT was associated with a 3.7% increase in MAFLD risk (OR = 1.037, 95% CI: 1.024–1.051, P < 0.001). Because univariate odds ratios reflect crude associations only, a multivariable logistic regression model including mean hepatic CT attenuation, VAT, SAT, liver volume, and age was additionally fitted to identify independent predictors. After mutual adjustment, mean hepatic CT attenuation (adjusted OR per 1-HU increase = 0.74, 95% CI: 0.65 to 0.84, P < 0.001) and VAT (adjusted OR per 1-cm³ increase = 1.025, 95% CI: 1.011 to 1.040, P < 0.001) remained independent predictors, whereas SAT and liver volume were no longer significant after adjustment, indicating partial collinearity with VAT and BMI-related variables. DeLong testing for two correlated ROC curves showed that the AUC of the combined model (0.928) did not reach a statistically significant difference compared with that of mean hepatic CT attenuation alone (0.923; ΔAUC = 0.005, P = 0.62).

Fig. 3.

Forest plot of univariate logistic regression. Squares indicate odds ratios; red = P < 0.05, gray = not significant

Therefore, the increase in AUC was not statistically significant. The gain in sensitivity from 85.8% to 91.7% should be interpreted as a clinically modest improvement rather than a confirmed superior performance. Bootstrap internal validation (1000 resamples) of the combined model yielded an optimism-corrected AUC of 0.912 and a calibration slope of 0.94, consistent with limited but acceptable overfitting in this single-center cohort.

ROC analysis (Fig. 4) is summarized in Table 3. Mean hepatic CT attenuation had the highest AUC (0.923) with an optimal cutoff of 48.7 HU. The combined model yielded an AUC of 0.928 and a sensitivity of 91.7%. A side-by-side comparison of AUC, sensitivity, and specificity across all CT parameters is presented in Fig. 5.

Fig. 4.

ROC curves. A Individual CT parameters; (B) Combined model vs. mean hepatic CT attenuation

Table 3.

Diagnostic performance of CT parameters for MAFLD

Diagnostic Parameter	AUC (95% CI)	Cutoff Value	Sensitivity (%)	Specificity (%)
Mean hepatic CT attenuation (HU)	0.923 (0.879–0.967)	48.7	85.8	87.0
Mean CT attenuation of L and R lobes (HU)	0.886 (0.833–0.939)	54.0	85.0	77.8
Liver volume (cm³)	0.814 (0.749–0.879)	1429	76.7	75.9
VAT (cm³)	0.796 (0.726–0.866)	109	54.2	90.7
SAT (cm²)	0.735 (0.656–0.814)	63	83.3	57.4
VAT/SAT ratio	0.601 (0.510–0.692)	1.6	30.0	87.0
Combined model*	0.928 (0.886–0.970)	0.6	91.7	85.2

*The combined model included SAT, VAT, and mean hepatic CT attenuation. Cutoff values were determined by maximizing the Youden index

Fig. 5.

Bar chart comparing AUC, sensitivity, and specificity across parameters

Density distribution and cutoff values

Kernel density plots (Fig. 6) depict the distributional variations in each CT parameter across the groups. The clearest differentiation was observed for mean hepatic CT attenuation, with a threshold of 48.7 HU effectively separating the two cohorts.

Fig. 6.

Kernel density plots showing parameter distribution in the two groups. (A) Liver mean CT value (HU); (B) Visceral adipose tissue volume (VAT, cm³); (C) Subcutaneous adipose tissue area (SAT, cm²); (D) Liver volume (cm³)

Combined model validation

The waterfall chart (Fig. 7A) demonstrates that the majority of patients were accurately classified by the model, achieving a total accuracy rate of 87.9%. The calibration plot (Fig. 7B) verified a strong concordance between the predicted and the actual observed probabilities.

Fig. 7.

A. Waterfall plot of predicted probabilities. B. Calibration curve of the combined model

Agreement between left and right hepatic lobe CT attenuation

Bland–Altman analysis (Fig. 8A) demonstrated a mean difference of − 0.31 HU between the left and right lobes (95% limits of agreement: −12.54 to 11.92 HU), with the vast majority of data points falling within these boundaries. The scatter plot (Fig. 8B) verified a robust linear correlation (r = 0.946, P < 0.001).

Fig. 8.

(A) Bland–Altman plot; (B) Scatter plot for left vs. right lobe CT attenuation

Summary of AUC rankings

Ranked by AUC (Fig. 9), the parameters were ordered as follows: combined model (0.928), mean hepatic CT attenuation (0.923), mean CT attenuation of the left and right lobes (0.886), liver volume (0.814), VAT (0.796), SAT (0.735), and VAT/SAT ratio (0.601).

Fig. 9.

AUC ranking of CT parameters for MAFLD diagnosis

Discussion

Diagnostic value of hepatic CT attenuation

The MAFLD group showed optimal diagnostic performance in terms of mean hepatic CT attenuation (AUC = 0.923; cutoff 48.7 HU; sensitivity 85.8%); specificity (87.0%). Compared with the controls, MAFLD showed significantly lower hepatic CT attenuation values (P < 0.001).

Pickhardt et al. [16] reported a strong linear relationship of unenhanced CT hepatic attenuation to MRI-PDFF(r²=0.83) in a multicenter study, and noted that lower than 48 CT attenuation values are reduced by the accumulation of triglyceride in hepatic cells, which is pathologically related to X-ray attenuation because fat tissue has a weaker capacity for photon absorption compared with normal liver parenchyma; hence higher fat content means a lower CT density. This 48.7 HU threshold (slightly below common 55–65 HU liver CT attenuation values) has been validated in clinical practice by Xu’s study [17], whose outcome is very similar to ours, showing intermediate steatosis in a Chinese MAFLD population. The high level of agreement between the quantitative CT and MRI-Dixon-quant estimates for liver fat content in Chinese subjects, as shown in [18], can support CT as an alternative to MRI. Long et al. [19] demonstrated that CT attenuation values from non-heavy drinkers are reliable indicators of the degree of hepatic steatosis. Demonstrating that mean hepatic CT attenuation is a straightforward, repeatable procedure that requires no special software and is therefore well suited to ad hoc evaluation during standard abdominal CT examinations. Taken together these data indicate that mean hepatic CT attenuation can be considered as reliable marker for MAFLD and the cutoff value approximates 48 HU.

Visceral versus subcutaneous fat: comparative diagnostic value

The diagnostic accuracy of VAT in MAFLD was higher than that of SAT, and VAT / SAT had the lowest efficacy as a diagnostic marker (AUC = 0.601). Therefore, it suggests that absolute measurement is better than relative measurement when using CT imaging to assess MAFLD. The superior performance of VAT is consistent with the metabolic role of visceral adipose tissue. Van der Poorten et al.’s [20] study showed that visceral fat plays a critical role in steatohepatitis, as free fatty acids from visceral stores are released directly into the liver through the portal vein and so accelerate fat accumulation. In a Chinese cohort, visceral tissue exhibits greater lipolytic activity and secretion of proinflammatory cytokines such as TNF-α and IL-6, which subsequently enter the portal circulation towards the liver, thereby aggravating inflammation and steatosis compared with subcutaneous fat. Zhao [21] reported that there is good agreement between CT fat measures in the abdomen and NAFLD score; meanwhile, Shi et al. [22] also found a significant association between distribution of abdominal fat measured on CT images and degree of hepatic steatosis.

Zhang et al. [23] established the optimal diagnostic cutoff values of quantitative CT fat indicators in MS, and again, VAT was superior to SAT; this finding was also demonstrated by showing correlation of hepatic/splenic CT value ratio to atherosclerotic plaque in fatty liver disease, which indicates a greater potential utility of measuring CT fat for patients’ outcomes. It is noteworthy that the VAT/SAT ratio did not demonstrate substantial improvement. Two reasons may be attributable to this observation: the MAFLD group, who were mainly overweight and/or obese, exhibited elevated levels of both VAT and SAT; and ratio-based indices tend not to incorporate absolute quantities. Future studies that include more normal-weight participants may further clarify the clinical utility of ratio indices.

CT–Metabolic index correlations and clinical implications

The complexity of MAFLD as a multifactorial metabolic disorder is supported by correlations between hepatic CT attenuation with body mass index (r=-0.607), TG(r = -0.418), ALT(r = -0.381), and GGT(r = -0.342), while VAT has a positive correlation with BMI(r = 0.554)and UA(r = 0.362). The negative linear relationship between hepatic CT attenuation values and hepatic TG content implies a degree of correspondence in severity between fatty liver grade and hepatocyte injury, which relates to the underlying pathology of hepatic fat accumulation, which was also shown correspondingly through the negative associations with ALT and GGT; lower CT value indicates more fat infiltration, greater inflammation. It is noteworthy that there appears to be a link between visceral fat and UA, since high levels of UA exacerbate liver steatosis through increased oxidative and ER stress, while visceral fat-induced IR hampers the kidney’s ability to excrete urates and hence forms a vicious cycle – Liang et al. [24]. Similar correlations were found in a Chinese population of CT-measured steatosis with other metabolic parameters by one group [22], and the correlation between visceral fat volume on abdominal CT scans and circulating adipokines and NAFLD [25], thus confirming that CT fat measures are useful in metabolic phenotyping. These correlations suggest that CT measurements could contribute to diagnosing MAFLD and assessing the global metabolic status of patients. In light of recommendations made in 2024 [8]for a detailed metabolic workup of patients with MAFLD, CT-derived parameters may contribute to such an approach.

Combined diagnostic model

The combined model, including SAT, VAT, and the mean hepatic CT attenuation, resulted in an AUC value of 0.928 at a sensitivity level of 91.7% and a specificity level of 85.2%, which is marginally higher than using hepatic CT attenuation alone (AUC = 0.923). The calibration plot showed good agreement between prediction and observation. Using hepatic density in conjunction with fat distribution parameters, the model can address complementary aspects of disease pathophysiology. Thus, by quantifying steatosis in the liver using hepatic CT attenuation, SAT and VAT measurement may be used to provide information about system-wide adipose tissue distribution. However, suggesting that the sole use of hepatic CT value would miss some patients with early stage and/or heterogeneous density by using fat distribution parameters. The higher sensitivity of the combined model(91.7%) compared to hepatic CT attenuation alone (85.8%) clearly suggests a potential advantage for minimizing missed cases, although this increase should be interpreted with caution given the non significant DeLong test. Ultrasound-based CAP, however, has an AUC range of 0.77–0.82 and is less accurate in the obese population. Our CT-based model (AUC = 0.928), conversely, demonstrated favorable diagnostic performance in this cohort. This integrated model with comparable AUC to MRI-PDFF could serve as a complementary examination when routine abdominal CT is performed, thus permitting concurrent screening for MAFLD and other abdominal pathologies. Of note, the increment in AUC from mean hepatic CT attenuation alone (0.923) to the combined model (0.928) was small and did not reach statistical significance on the DeLong test (ΔAUC = 0.005, P = 0.62). The incremental value of the combined model over mean hepatic CT attenuation alone was not statistically significant; therefore, we refrain from claiming superiority and instead emphasize its complementary role. The value of incorporating SAT and VAT lies less in raising overall discrimination than in providing complementary information on adipose-tissue distribution, which may offer a favorable impact on sensitivity for early or heterogeneous steatosis and supports global metabolic phenotyping. Comparisons with CAP and MRI-PDFF in this paragraph were drawn from heterogeneous published cohorts and should be interpreted as descriptive rather than head-to-head benchmarks, given that a direct comparison within the same patients was not performed in our study.

Left–right hepatic lobe agreement

Bland-Altman analysis indicated minimal bias (-0.31 HU) in CT attenuations for the left and right hepatic lobe, with the 95% agreement (from − 12.54 to 11.92 HU) and high correlation coefficient (r = 0.946). This degree of agreement indicates that in cases where there are focal lesions affecting the contralateral lobe or postsurgery changes, or no space to measure, measuring both lobes is feasible. However, in some cases, the differences between lobes exceeded 10 HU, which can indicate the asymmetric hepatic steatosis. We recommend the common approach of measuring each lobe and taking an average to account for local variation in fat distribution, as has been performed by Li et al. [26]; we consider it more robust than using only one lobe measure for individual subjects (which might be inaccurate).reinforcing the importance of keeping a fixed location of regions of interest and averaging across sites to increase the confidence in measurements, who advocated the same approach.

Limitations and future directions

The generalizability of our results is limited because we had a relatively modest sample size (n = 174), including 54 control subjects. Because this study was a single-center, retrospectively designed study, there might have been some degree of selection bias; therefore, certain limitations are noted. The absence of a gold standard histology in the form of a liver biopsy led to the diagnosis of MAFLD being made using imaging and clinical criteria only. Additionally, we excluded conventional CT parameters such as liver/spleen attenuation ratio from our comparison, and finally, the joint prediction model must be validated in an external independent cohort.

Further research will need to be done in the future using large-scale prospective multicentre studies with a heterogeneous population for accurate validation of this model. A reference standard, such as liver biopsy or MRI-PDFF, can facilitate quantification of the relationship between CT values and levels of steatosis. Deep learning-based automatic CT image segmentation and fat quantification may enhance efficiency and standardization in measurements.

It would be valuable to develop a multivariable model integrating other CT markers with laboratory markers. While recommending that those with the features of the metabolic syndrome undergo screening for NAFLD, the 2025 MAFLD primary care guidelines [27] suggest quantitative CT-based evaluation may be a useful adjunct to US screening, thus paving the way toward more accurate noninvasive methods of diagnosing MAFLD.

Several additional limitations require explicit acknowledgement and temper the strength of our conclusions. First, the reference standard for MAFLD in this study relied on imaging evidence of hepatic steatosis combined with metabolic criteria, while CT-derived parameters were simultaneously evaluated as index tests; this produces an inherent incorporation bias, because hepatic CT attenuation is part of the diagnostic definition and is then reused as a candidate predictor. The reported AUC for mean hepatic CT attenuation should therefore be interpreted with caution, and the comparative diagnostic accuracy of CT parameters can only be definitively established against an independent reference such as MRI-PDFF or liver biopsy. Second, the MAFLD group was significantly younger than the non-MAFLD group (48.2 vs. 54.5 years), and although a sensitivity analysis in an age-stratified subset supported robustness, residual confounding by age and age-related differences in body composition cannot be fully excluded; future work should rely on age- and sex-matched controls or prospective enrollment with prespecified matching. Third, we conducted only internal bootstrap validation; the combined model has not yet been validated in an external, geographically and demographically distinct cohort, and the modest non-significant ΔAUC (0.005) over hepatic attenuation alone (DeLong P = 0.62) underscores that the model’s incremental value should be regarded as hypothesis-generating until externally confirmed. Fourth, conventional ratio-based indices, in particular the liver-to-spleen CT attenuation ratio and the liver minus spleen HU difference (ΔHU), were not included in our primary comparison; although these parameters were recorded for every patient, their formal comparative evaluation against the combined model is a planned analysis in our ongoing prospective extension. Finally, regarding radiation dose, all scans were clinically indicated unenhanced abdominal CT examinations acquired under automatic tube current modulation, so no additional radiation was incurred for research purposes; the dose-length product distribution was consistent with national diagnostic reference levels, and the opportunistic use of these scans does not impose any incremental radiation burden on patients. Given the single-center retrospective design, the modest sample size, and the reliance on an imaging-based reference standard, the clinical interpretation of our findings should be appropriately cautious: the combined model is best regarded as a promising opportunistic screening tool that requires prospective, multicenter validation rather than a confirmed alternative to established quantitative methods such as MRI-PDFF.

Conclusion

The mean liver CT attenuation was the most effective single index in CT-measured abdominal fat indices to diagnose MAFLD. More accurate results could be obtained when employing the composite model consisting of SAT, VAT, and liver attenuation (AUC = 0.928; sensitivity = 91.7%; specificity = 85.2%). The diagnostic value of VAT was higher than that of SAT, consistent with the fact that it plays an indispensable role as a pathogenic factor in MAFLD. The CT parameters are correlated significantly with metabolic indicators, which suggests that they can be used to reflect general metabolic healthQuantitative fat assessment represents a feasible strategy in the screening of incidental MAFLD.

Authors' contributions

Jun Wang: Conceptualization, methodology, data curation, formal analysis, investigation, writing—original draft. Hongyan Wang: Investigation, data curation, visualization, writing—review and editing. Jianliang Wang: Investigation, validation, resources, data curation. Jie Feng: Formal analysis, methodology, visualization, writing—review and editing. Zhen Zhang: Conceptualization, supervision, project administration, funding acquisition, writing—review and editing. All authors have read and approved the final manuscript.

Funding

The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.

Data availability

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

Declarations

Ethics approval and consent to participate

This study was performed in accordance with the ethical standards of the institutional research committee and with the 1964 Helsinki Declaration and its later amendments. The study was approved by the institutional ethics committee of the Affiliated Hospital of Jiangsu University. Informed consent was waived by the Institutional Ethics Committee of the Affiliated Hospital of Jiangsu University due to the retrospective nature of the study and the use of anonymized data.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.