Validation of voxel-based computed tomography liver fat quantification for assessing incident type 2 diabetes risk in a health check-up population

Introduction

Diabetes mellitus is a major chronic metabolic disease worldwide, with a steadily increasing incidence that poses a substantial burden on public health (1-3). Metabolic dysfunction-associated fatty liver disease (MAFLD) is a form of liver disease closely related to metabolic abnormalities and is particularly common in the setting of obesity, type 2 diabetes (T2D), and insulin resistance. The definition of MAFLD emphasizes the role of metabolic dysfunction in the development of hepatic steatosis and highlights its close association with metabolic disorders such as diabetes (2). There is a well-established bidirectional relationship between MAFLD and T2D, with individuals with MAFLD having a markedly increased risk of developing diabetes (4). This interrelationship not only influences disease progression but also exacerbates cardiovascular and liver-related complications. Therefore, accurate and effective quantification of hepatic fat burden is essential for understanding the pathophysiology of MAFLD and for predicting the risk of diabetes.

Among the available imaging modalities, magnetic resonance imaging (MRI), particularly chemical shift-encoded MRI and proton density fat fraction (PDFF) mapping, is regarded as the most accurate and precise noninvasive method for liver fat quantification, providing volumetric fat maps with high reproducibility and strong correlation with the histologic fat content (5-8). However, the routine use of MRI is limited by the scanning time, cost, and availability. In contrast, computed tomography (CT) is widely performed in abdominal imaging and is commonly included in routine health check-ups, allowing the incidental detection of hepatic steatosis. Traditional CT-based approaches to assess liver fat rely on attenuation criteria such as absolute Hounsfield unit thresholds or liver-to-spleen ratios, and in clinical practice, these assessments are often based on simple visual or manual interpretation rather than automated whole-organ quantification, making them susceptible to imaging parameter variation (5,9-12).

With the increasing emphasis on early identification of metabolic risk, accurate liver fat quantification has become particularly important in apparently healthy populations undergoing routine imaging, in whom metabolic dysfunction may not yet be clinically evident. Although CT inherently provides voxel-level attenuation information, precise voxel-based quantification of hepatic fat has only recently become feasible (6,7,9,11,13). A novel voxel-based CT method has been proposed to enable automated whole-organ quantification of liver fat from routine CT images, potentially overcoming the limitations of conventional CT-based assessments (14). However, the clinical utility of this CT-derived voxel-based liver fat quantification for predicting future T2D and assessing diabetes risk in a healthy population remains unclear.

In this study, we aimed to evaluate whether CT-derived voxel-based quantification of liver fat can predict the development of T2D and assess future diabetes risk in a health check-up population. We present this article in accordance with the TRIPOD reporting checklist (available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1-0342/rc).

Methods

Study population

This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study protocol was approved by the Institutional Review Board of The Second People’s Hospital of Bengbu (approval number: 2023-288), and the requirement for informed consent was waived because of the retrospective nature of the study.

This retrospective cohort study included individuals who underwent routine health check-up abdominal CT examinations at The Second People’s Hospital of Bengbu between January 2013 and December 2023. The study population was derived from a longitudinal health check-up database that integrates imaging, clinical, laboratory, and follow-up medical records. The baseline was defined as the date of the first eligible abdominal CT examination during the study period.

Individuals were eligible for inclusion if they underwent noncontrast abdominal CT as part of a routine health check-up and had no evidence of T2D at baseline. Participants were excluded if they met any of the following criteria: (I) prevalent T2D at baseline; (II) follow-up duration of less than 1 year or only a single health check-up record; (III) missing baseline laboratory data; (IV) missing CT data or poor image quality; (V) a history of chronic liver disease (viral hepatitis or cirrhosis) or hepatic malignancy; or (VI) prior major liver surgery affecting the liver (Figure 1A).

Figure 1 Study population and representative example of voxel-based liver fat quantification. (A) Flowchart of study population selection. (B) Representative example of CT-based voxel-based liver fat identification. The liver segmentation is shown in brown, and fat-containing voxels identified by the algorithm are shown in yellow. CT, computed tomography.

Outcome and follow-up

The primary outcome was incident T2D during follow-up. Diabetes status was ascertained through retrospective review of outpatient, inpatient, and health check-up records. Incident T2D was defined on the basis of documented clinical diagnosis by a physician, in accordance with routine clinical practice.

Participants were followed from the date of the baseline CT examination until the occurrence of incident T2D, the date of the last available medical record, or June 2025, whichever occurred first.

Voxel-based liver fat quantification

The present study aimed to evaluate whether a voxel-based quantification method for metabolic-associated fatty liver on CT images, as previously described (14), could predict the future risk of incident T2D in a health check-up population.

Liver fat was quantified on baseline noncontrast abdominal CT images using a previously described voxel-based CT framework (14). For each examination, the CT image and the corresponding multiorgan segmentation mask were loaded. The liver and spleen were segmented via TotalSegmentator (v2.5.0) (15), and organ regions were identified on the basis of predefined segmentation labels.

To assess the reliability of automated organ segmentation, we randomly selected 36 participants (10% of the final cohort), and a board-certified radiologist (H.L.) manually delineated three-dimensional liver and spleen regions of interest on CT images. The Dice similarity coefficients between the automated TotalSegmentator-derived masks and the manual regions of interest were 0.964±0.014 [mean ± standard deviation (SD)] for the liver and 0.983±0.015 for the spleen.

To reduce the influence of extreme attenuation values, the CT intensities were clipped to the 2.5^th and 97.5^th percentiles within the combined liver-spleen region. The mean spleen attenuation was then calculated from all voxels within the spleen mask and was used as the reference value for the spleen-based criteria. Each liver voxel was then evaluated according to three attenuation-based criteria (14): (I) absolute liver voxel attenuation <40 HU; (II) liver voxel attenuation divided by mean spleen attenuation <1; and (III) liver voxel attenuation minus mean spleen attenuation ≤−10 HU. A liver voxel was classified as fat-containing if at least two of these three criteria were satisfied.

On the basis of this voxel wise classification, a binary liver fat mask was generated for each scan. The average liver fat fraction was defined as the proportion of liver voxels classified as fat and was used as the primary imaging biomarker in subsequent analyses.

Figure 1B provides an example of the proposed liver fat quantification method. In addition, we calculated the mean liver attenuation.

Baseline clinical and laboratory assessment

Baseline clinical characteristics and laboratory measurements were obtained from the health check-up visit closest to the baseline CT examination. Demographic variables and routine laboratory tests, including demographic variables (age and sex), anthropometric measurements (body mass index), glycemic indices [hemoglobin A1c (HbA1c)], lipid profiles [total cholesterol, low-density lipoprotein (LDL) cholesterol, high-density lipoprotein (HDL) cholesterol, and triglycerides], liver enzymes [aspartate aminotransferase (AST), alanine aminotransferase (ALT), and γ-glutamyltransferase], and renal function parameters (serum creatinine), were extracted from electronic medical records. Laboratory measurements were generally performed on the same day as the CT examination.

Statistical analysis

Continuous variables are presented as medians [interquartile ranges (IQRs)], and categorical variables are presented as numbers (percentages). Comparisons between groups were performed via the Wilcoxon rank-sum test for continuous variables and the χ² test for categorical variables. All the statistical tests were two-sided, and a P value <0.05 was considered statistically significant unless otherwise specified. All analyses were performed using a complete-case approach.

The association between liver fat and the risk of incident T2D was evaluated via Cox proportional hazards regression models. The average liver fat fraction was standardized as a z-score before Cox regression analysis to facilitate interpretation of the hazard ratio (HR) as the risk associated with each 1-SD increase. Univariable Cox regression analyses were first performed to screen potential clinical predictors and the CT-derived average liver fat fraction. Variables with a P value <0.10 in the univariable analyses were subsequently entered into the multivariable Cox model. HRs and 95% confidence intervals (CIs) are reported.

To explore potential nonlinear relationships between the liver fat fraction and incident T2D incidence, restricted cubic spline (RCS) functions were incorporated into the Cox models. The spline model was fitted via four knots located at the 5^th, 35^th, 65^th, and 95^th percentiles of the liver fat fraction distribution, with the median value used as the reference. P values for overall association and nonlinearity were calculated.

The predictive performance of the liver fat fraction for incident T2D was assessed via time-dependent receiver operating characteristic (ROC) analysis at 3 and 5 years. Time-dependent area under the curve (tAUC) values were estimated via an inverse probability of censoring weighting (IPCW) approach.

Model calibration at 3 and 5 years was evaluated via calibration plots, which were used to compare the predicted risks with the observed event rates. Decision curve analysis (DCA) was performed to assess the potential clinical utility of liver fat-based prediction models across a range of clinically relevant threshold probabilities.

For risk stratification, participants were grouped into low- and high-risk groups on the basis of the median value of the liver fat fraction. The cumulative incidence of T2D was estimated via the Kaplan‒Meier method, and differences between groups were assessed via the log-rank test. HRs comparing high- vs. low-risk groups were derived from Cox proportional hazards models.

To evaluate the incremental predictive value of liver fat beyond clinical risk factors, multiple Cox-based prediction models were constructed and compared. First, Cox models based on HbA1c alone, liver fat alone, and liver fat combined with HbA1c were developed. Second, a clinical prediction model was derived using least absolute shrinkage and selection operator (LASSO) Cox regression. Candidate variables for the LASSO Cox model included demographic, anthropometric, glycemic, lipid, liver enzyme, blood pressure, renal function, and insulin-related variables, including age, sex, body mass index, HbA1c, fasting glucose, triglycerides, HDL cholesterol, LDL cholesterol, total cholesterol, ALT, AST, gamma-glutamyl transferase (GGT), systolic blood pressure, diastolic blood pressure, uric acid, creatinine, and fasting insulin. The liver fat fraction was not included during LASSO-based clinical variable selection, so that its incremental predictive value could be evaluated separately. Finally, the liver fat fraction was added to the LASSO-derived clinical Cox model, and model performance was compared using the C-index and 3- and 5-year time-dependent AUCs. All the statistical analyses were performed via Python (version 3.10).

Results

Study population

Between January 2013 and December 2023, a total of 3,018 individuals underwent routine health check-up abdominal CT examinations. After applying the exclusion criteria, 356 participants were included in the final analysis (Figure 1A). The median age of the study population was 57 years (IQR, 53–63 years), and 201 (56.46%) participants were male. During a median follow-up of 4.5 years (IQR, 3.1–7.1), 32 participants (9.0%) developed incident T2D, whereas 324 participants remained free of T2D. The baseline characteristics of the study population stratified by incident T2D status are summarized in Table 1.

Association between liver fat and incident T2D

According to the univariable Cox regression analysis, a greater average liver fat fraction was significantly associated with an increased risk of incident T2D (HR per SD, 2.11; 95% CI: 1.43–3.12; P<0.001). After adjustment for demographic factors, body mass index, glycemic indices, including HbA1c, lipid profiles, liver enzymes, blood pressure, and renal function parameters, the average liver fat fraction remained independently associated with incident T2D (HR per SD, 1.65; 95% CI: 1.05–2.60; P=0.031) (Table 2). In contrast, the association between HbA1c and incident T2D was attenuated after multivariable adjustment (HR, 1.62; 95% CI: 0.66–3.96; P=0.288). We performed a sensitivity analysis adjusted for clinically established diabetes risk factors, including age, sex, body mass index, and HbA1c. In this model, the average liver fat fraction remained significantly associated with incident T2D (HR per SD, 1.89; 95% CI: 1.26–2.83; P=0.002).

RCS analysis demonstrated a monotonic increase in T2D risk with increasing average liver fat fraction, with a significant overall association (P for overall <0.001) and no strong evidence of nonlinearity (P for nonlinearity =0.528) (Figure 2).

Figure 2 Restricted cubic spline analysis of average liver fat fraction and incident T2D. The association between the average liver fat fraction and the risk of incident T2D was evaluated via a Cox proportional hazards model with a restricted cubic spline function (four degrees of freedom). HRs are expressed relative to the median value of the average liver fat fraction (vertical dashed line), with shaded areas indicating 95% CIs. The model was adjusted for baseline age, sex, BMI, and HbA1c. P for overall fat fraction was derived from a likelihood ratio test comparing models with and without spline terms for average liver fat fraction, whereas P for nonlinearity was obtained by comparing the spline model with a model including average liver fat fraction as a linear term only. BMI, body mass index; CI, confidence interval; HbA1c, hemoglobin A1c; HR, hazard ratio; T2D, type 2 diabetes.

Predictive performance of liver fat for incident T2D

The predictive performance of the average liver fat fraction for incident T2D was evaluated via time-dependent discrimination, calibration, clinical utility, and risk stratification analyses (Figure 3). Time-dependent ROC analysis demonstrated moderate discrimination, with tAUCs of 0.71 (95% CI: 0.59–0.84) at 3 years and 0.72 (95% CI: 0.60–0.82) at 5 years.

Figure 3 Performance and clinical utility of the average liver fat fraction for predicting incident T2D. (A) Time-dependent ROC curves at 3 and 5 years (3y and 5y). (B) Calibration plots at 3 and 5 years comparing the predicted and observed risks. (C) Decision curve analysis showing net benefit across clinically relevant threshold probabilities. (D) Cumulative incidence of incident T2D stratified by the median split of the average liver fat fraction, with shaded areas indicating 95% CIs. HRs were derived from Cox proportional hazards models, and the log-rank P value was added to compare cumulative incidence curves between groups. The risk table shows the number of participants at risk at each follow-up time point. AUC, area under the curve; CI, confidence interval; DCA, decision curve analysis; HR, hazard ratio; ROC, receiver operating characteristic; T2D, type 2 diabetes; y, year.

The calibration plots at 3 and 5 years showed reasonable agreement between the predicted and observed risks across risk strata. DCA indicated that liver fat-based models provided a greater net benefit than did treat-all or treat-none strategies across a range of clinically relevant threshold probabilities for both 3- and 5-year risk horizons.

For risk stratification, participants were dichotomized into low- and high-risk groups according to the median average liver fat fraction. Individuals in the high-liver fat group presented a significantly greater cumulative incidence of T2D than did those in the low-liver fat group did (HR, 2.71; 95% CI: 1.26–5.87), supporting the clinical relevance of liver fat-based risk classification.

Predictive performance of liver fat for incident T2D

To assess the incremental predictive value of liver fat for incident T2D, we compared liver fat-based models with HbA1c-based and clinical models. Compared with HbA1c alone, the combination of liver fat and HbA1c further improved the predictive performance, indicating that liver fat provides prognostic information beyond glycemic status. A clinical model based on demographic and metabolic variables achieved moderate discrimination. In a clinically relevant setting, adding liver fat to the LASSO-derived clinical model resulted in the highest overall discrimination across both the 3- and 5-year prediction horizons, supporting the incremental value of CT-derived liver fat beyond established clinical risk factors (Table 3). Moreover, the proposed method outperformed conventional liver attenuation in predicting the risk of incident T2D.

Discussion

In this retrospective cohort study of individuals undergoing routine health check-up abdominal CT examinations, we performed a prognostic validation of a previously proposed CT-derived voxel-based average liver fat fraction for incident T2D risk assessment. We demonstrated that this CT-derived voxel-based liver fat fraction was significantly associated with the future development of T2D. This association remained robust after adjustment for established metabolic risk factors, including HbA1c and body mass index, suggesting that voxel-based liver fat quantification captures metabolic risk information beyond conventional glycemic and anthropometric measures.

In addition to its association with disease risk, the voxel-based average liver fat fraction showed moderate and consistent predictive performance for incident T2D across multiple evaluation domains. Time-dependent ROC analyses demonstrated stable discrimination at both 3 and 5 years, whereas calibration analyses indicated reasonable agreement between the predicted and observed risks. DCA further suggested that liver fat-based models provided meaningful clinical net benefit across a range of clinically relevant threshold probabilities. Importantly, risk stratification on the basis of the median liver fat fraction identified a subgroup of individuals with substantially greater cumulative incidence of T2D, highlighting the potential utility of voxel-based liver fat quantification for identifying high-risk individuals in routine health check-up settings.

Recent studies have demonstrated the feasibility of automated CT-based body composition analysis in large-scale cohorts, including automated segmentation of multiple organs and tissue compartments as well as voxel-wise body composition analysis based on image registration. These developments support the broader use of CT-derived quantitative imaging biomarkers in metabolic risk assessment (16,17). In this context, our findings extend previous imaging-based studies of hepatic steatosis by demonstrating the prognostic value of a voxel-based CT-derived liver fat metric in a metabolically heterogeneous but apparently healthy population (14,18). Traditional CT-based assessments of liver fat, which rely on mean attenuation values or liver-to-spleen ratios, have primarily been used for qualitative or semiquantitative evaluation and are often influenced by imaging parameters and manual interpretation (5,16). In contrast, the voxel-based approach applied in this study leverages the inherent voxel-level information in CT images to achieve automated whole-organ quantification of hepatic fat. Consistent with this, we observed that the voxel-based average liver fat fraction outperformed conventional mean liver attenuation in predicting incident T2D, suggesting that voxel-level analysis may better capture the extent and heterogeneity of fat-containing regions within the liver (19-21).

From a clinical perspective, the incremental predictive analyses further support the relevance of voxel-based liver fat quantification. Compared with HbA1c alone, liver fat alone demonstrated better discrimination, and combining liver fat with HbA1c improved the predictive performance compared with the use of either marker individually (22). While a clinical model derived from demographic and metabolic variables achieved moderate discrimination, the addition of voxel-based liver fat resulted in the highest overall predictive performance across both the 3- and 5-year horizons. These findings suggest that CT-derived voxel-based liver fat provides complementary prognostic information beyond established clinical risk factors and may enhance risk assessment strategies for T2D in populations undergoing routine imaging (22-25).

Several limitations of this study should be acknowledged. First, this was a retrospective, single-center study, which may limit the generalizability of the findings to other populations, institutions, and imaging settings. In addition, although the study population was derived from routine health check-up participants, it may not fully represent the broader health check-up population because all participants underwent abdominal CT, had complete baseline laboratory data, and had available longitudinal medical records. Second, the number of incident T2D events was relatively small, which may have limited the statistical power of the analyses and the stability of the multivariable and predictive models. Therefore, the present findings should be interpreted cautiously and require confirmation in larger cohorts. Third, incident T2D was identified through a review of outpatient, inpatient, and health check-up records; therefore, undiagnosed, subclinical, or externally diagnosed diabetes cases may have been missed. Fourth, residual confounding cannot be fully excluded. In particular, detailed information on diet and lifestyle factors, such as dietary pattern, physical activity, alcohol intake, and smoking status, was not systematically available because of the retrospective study design. These factors may influence both liver fat accumulation and diabetes risk and could therefore modify the observed association. Fifth, liver fat quantification was derived from routine CT examinations with varying acquisition parameters, which may have introduced measurement variability despite the use of standardized voxel-based criteria. Moreover, the average liver fat fraction used in this study represents an algorithm-derived voxel fraction rather than a direct measure of absolute hepatic fat content; therefore, clinically meaningful cut-offs cannot be directly inferred from conventional liver fat percentage thresholds. Future multicenter studies are needed to validate this CT-derived voxel-based metric in broader populations, calibrate it against established reference standards, and define clinically meaningful cut-offs for risk stratification.

Conclusions

In conclusion, voxel-based CT quantification of the average liver fat fraction was independently associated with the future development of T2D in a health check-up population. Compared with conventional CT attenuation measures and established clinical risk factors, voxel-based liver fat measurements may provide additional prognostic information for diabetes risk assessment. These findings suggest that automated voxel-based liver fat quantification from routine CT images shows potential as a practical imaging biomarker for identifying individuals at increased risk of T2D. Further multicenter studies with larger and diverse populations are warranted to validate this CT-derived voxel-based metric, calibrate it against established reference standards, and define clinically meaningful cut-offs for risk stratification and potential clinical implementation.

Acknowledgments

During the preparation of this work, the authors used ChatGPT 5 in order to improve the clarity and readability of the language. After using this service, the authors reviewed and edited the content as needed and takes full responsibility for the content of the published article.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1-0342/rc

Data Sharing Statement: Available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1-0342/dss

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://qims.amegroups.com/article/view/10.21037/qims-2026-1-0342/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study protocol was approved by the Institutional Review Board of The Second People’s Hospital of Bengbu (Approval Number: 2023-288), and the requirement for informed consent was waived because of the retrospective nature of the study.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Wong VW, Wong GL, Woo J, Abrigo JM, Chan CK, Shu SS, Leung JK, Chim AM, Kong AP, Lui GC, Chan HL, Chu WC. Impact of the New Definition of Metabolic Associated Fatty Liver Disease on the Epidemiology of the Disease. Clin Gastroenterol Hepatol 2021;19:2161-2171.e5. [Crossref] [PubMed]
2. Drożdż K, Nabrdalik K, Hajzler W, Kwiendacz H, Gumprecht J, Lip GYH. Metabolic-Associated Fatty Liver Disease (MAFLD), Diabetes, and Cardiovascular Disease: Associations with Fructose Metabolism and Gut Microbiota. Nutrients 2021;14:103. [Crossref] [PubMed]
3. Kaya E, Yilmaz Y. Metabolic-associated Fatty Liver Disease (MAFLD): A Multi-systemic Disease Beyond the Liver. J Clin Transl Hepatol 2022;10:329-38. [Crossref] [PubMed]
4. Qi X, Li J, Caussy C, Teng GJ, Loomba R. Epidemiology, screening, and co-management of type 2 diabetes mellitus and metabolic dysfunction-associated steatotic liver disease. Hepatology 2026;83:661-78. [Crossref] [PubMed]
5. Kramer H, Pickhardt PJ, Kliewer MA, Hernando D, Chen GH, Zagzebski JA, Reeder SB. Accuracy of Liver Fat Quantification With Advanced CT, MRI, and Ultrasound Techniques: Prospective Comparison With MR Spectroscopy. AJR Am J Roentgenol 2017;208:92-100. [Crossref] [PubMed]
6. Starekova J, Hernando D, Pickhardt PJ, Reeder SB. Quantification of Liver Fat Content with CT and MRI: State of the Art. Radiology 2021;301:250-62. [Crossref] [PubMed]
7. Damle V, Sharma A, Srivastava BK, Mohan V. Imaging Modalities for Hepatic Steatosis and Fibrosis in Type 2 Diabetes Mellitus: A Review Article. J Diabetol 2026;17:18-23.
8. Schwenzer NF, Springer F, Schraml C, Stefan N, Machann J, Schick F. Non-invasive assessment and quantification of liver steatosis by ultrasound, computed tomography and magnetic resonance. J Hepatol 2009;51:433-45. [Crossref] [PubMed]
9. Graffy PM, Sandfort V, Summers RM, Pickhardt PJ. Automated Liver Fat Quantification at Nonenhanced Abdominal CT for Population-based Steatosis Assessment. Radiology 2019;293:334-42. [Crossref] [PubMed]
10. Kodama Y, Ng CS, Wu TT, Ayers GD, Curley SA, Abdalla EK, Vauthey JN, Charnsangavej C. Comparison of CT methods for determining the fat content of the liver. AJR Am J Roentgenol 2007;188:1307-12. [Crossref] [PubMed]
11. Kim HN, Jeon HJ, Choi HG, Kwon IS, Rou WS, Lee JE, Lee TH, Kim SH, Lee BS, Shin KS, Lee HJ, Eun HS. CT-based Hounsfield unit values reflect the degree of steatohepatitis in patients with low-grade fatty liver disease. BMC Gastroenterol 2023;23:77. [Crossref] [PubMed]
12. Boyce CJ, Pickhardt PJ, Kim DH, Taylor AJ, Winter TC, Bruce RJ, Lindstrom MJ, Hinshaw JL. Hepatic steatosis (fatty liver disease) in asymptomatic adults identified by unenhanced low-dose CT. AJR Am J Roentgenol 2010;194:623-8. [Crossref] [PubMed]
13. Naghavi M, Atlas K, Reeves A, Zhang C, Wasserthal J, Atlas T, et al. AI-enabled opportunistic measurement of liver steatosis in coronary artery calcium scans predicts cardiovascular events and all-cause mortality: an AI-CVD study within the Multi-Ethnic Study of Atherosclerosis (MESA). BMJ Open Diabetes Res Care 2025;13:e004760. [Crossref] [PubMed]
14. Martín-Saladich Q, Pericàs JM, Ciudin A, Ramirez-Serra C, Escobar M, Rivera-Esteban J, Aguadé-Bruix S, González Ballester MA, Herance JR. Metabolic-associated fatty liver voxel-based quantification on CT images using a contrast adapted automatic tool. Med Image Anal 2024;95:103185. [Crossref] [PubMed]
15. Wasserthal J, Breit HC, Meyer MT, Pradella M, Hinck D, Sauter AW, Heye T, Boll DT, Cyriac J, Yang S, Bach M, Segeroth M. TotalSegmentator: Robust Segmentation of 104 Anatomic Structures in CT Images. Radiol Artif Intell 2023;5:e230024. [Crossref] [PubMed]
16. Ahmad N, Strand R, Sparresäter B, Tarai S, Lundström E, Bergström G, Ahlström H, Kullberg J. Automatic segmentation of large-scale CT image datasets for detailed body composition analysis. BMC Bioinformatics 2023;24:346. [Crossref] [PubMed]
17. Ahmad N, Dahlberg H, Jönsson H, Tarai S, Guggilla RK, Strand R, Lundström E, Bergström G, Ahlström H, Kullberg J. Voxel-wise body composition analysis using image registration of a three-slice CT imaging protocol: methodology and proof-of-concept studies. Biomed Eng Online 2024;23:42. [Crossref] [PubMed]
18. Tang Y, Wei ZM, Li N, Sun LL, Jin ZY, Wu Z, Sun H. Quantitative analysis of the risk of type 2 diabetes and fatty liver in non-obese individuals by computed tomography. Abdom Radiol (NY) 2022;47:2099-105. [Crossref] [PubMed]
19. Deng M, Li Z, Sun L, Wang H, Tang J, Liu Y, Chen X, Xu H, Dai Z. Liver MRI-derived Proton Density Fat Fraction Predicts Diabetes Risk in Patients with Obesity: A Dose-Response Study. Radiology 2025;317:e250026. [Crossref] [PubMed]
20. Zhang YX, Feng YP, You CL, Zhang LY. The diagnostic value of MRI-PDFF in hepatic steatosis of patients with metabolic dysfunction-associated steatotic liver disease: a systematic review and meta-analysis. BMC Gastroenterol 2025;25:451. [Crossref] [PubMed]
21. Reeder SB, Cruite I, Hamilton G, Sirlin CB. Quantitative Assessment of Liver Fat with Magnetic Resonance Imaging and Spectroscopy. J Magn Reson Imaging 2011;34:729-49. [Crossref] [PubMed]
22. Pickhardt PJ, Graffy PM, Reeder SB, Hernando D, Li K. Quantification of Liver Fat Content With Unenhanced MDCT: Phantom and Clinical Correlation With MRI Proton Density Fat Fraction. AJR Am J Roentgenol 2018;211:W151-7. [Crossref] [PubMed]
23. Brancato V, Della Pepa G, Bozzetto L, Vitale M, Annuzzi G, Basso L, Cavaliere C, Salvatore M, Rivellese AA, Monti S. Evaluation of a Whole-Liver Dixon-Based MRI Approach for Quantification of Liver Fat in Patients with Type 2 Diabetes Treated with Two Isocaloric Different Diets. Diagnostics (Basel) 2022;12:514. [Crossref] [PubMed]
24. Nasir M, Xu Y, Hasenstab K, Yechoor A, Dodhia R, Weeks WB, Ferres JL, Cunha GM. Liver MRI proton density fat fraction inference from contrast enhanced CT images using deep learning: A proof-of-concept study. PLoS One 2025;20:e0328867. [Crossref] [PubMed]
25. Naeem M, Bülow R, Schipf S, Werner N, Dörr M, Lerch MM, Kühn JP, Rathmann W, Nauck M, Markus MRP, Ittermann T, Völzke H. Association of hepatic steatosis derived from ultrasound and quantitative MRI with prediabetes in the general population. Sci Rep 2021;11:13276. [Crossref] [PubMed]