Association between low skeletal muscle mass assessed by deep learning-based CT and MASLD: a prospective cohort study

Abstract

Background

To explore the role of abdominal skeletal muscle mass in metabolic dysfunction-associated steatotic liver disease (MASLD) incidence and resolution, we established two independent cohorts in a Chinese population.

Methods

This study included 3139 participants without MASLD (incidence cohort) and 1130 MASLD patients (resolution cohort) at baseline, all of whom underwent abdominal computed tomography (CT) during routine health examinations from 2019 to 2024. Skeletal muscle mass was quantified from the third lumbar vertebra (L3) CT images using automated deep learning and analyzed in tertiles. MASLD was diagnosed by ultrasonographic hepatic steatosis and cardiometabolic risk factors. Cox models assessed the association of muscle mass with MASLD; logistic regression was used to assess the joint effects of skeletal muscle change rate and MASLD-related genetic risk on resolution.

Results

Among the incidence cohort (median follow-up: 1.8 years), 516 developed MASLD. Lower body weight-adjusted total abdominal muscle area (TAMA/weight) was associated with higher MASLD incidence, regardless of sex. Compared with the highest tertile of TAMA/weight, the lowest tertile had hazard ratio (HR) of 2.39 (95% confidence interval [CI], 1.75 to 3.26) in men, and 2.11 (95% CI, 1.45 to 3.07) in women, respectively. Higher annual TAMA/weight promoted MASLD resolution, with the odds ratio (OR) of 1.81 (95% CI, 1.28–2.56) in men and 1.75 (95% CI, 1.04–2.96) in women. Increased muscle mass tended to be more associated with MASLD resolution in low genetic risk individuals.

Conclusions

Lower muscle mass correlates with higher MASLD risk. Increasing skeletal muscle mass may help prevent and manage the disease.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13098-026-02279-w.

Introduction

The new term metabolic dysfunction-associated steatotic liver disease (MASLD) was formally proposed in 2023 to replace nonalcoholic fatty liver disease (NAFLD) [1, 2]. This change emphasizes the role of metabolism-related factors in the pathogenesis of fatty liver, not just co-existing with other liver diseases, which is more in line with the changes of modern disease spectrum. Given its strong connections to diabetes [3], obesity [4], and hypertension [5]—diseases whose prevalence has surged recently, MASLD prevalence among adults is projected to exceed 55% by 2040 [6]. MASLD is associated with an increased risk of hepatic complications, including cirrhosis, end-stage liver disease, and hepatocellular carcinoma and extrahepatic conditions such as cardiovascular disease, chronic kidney disease, and certain extrahepatic cancers [7, 8]. Consequently, it is essential to prevent and manage MASLD. Obesity, insulin resistance, unhealthy diet and older age are the most common risk factors for MASLD [9, 10]. However, low skeletal muscle mass has been identified in recent years [11].

Skeletal muscle is a major metabolic organ that regulates whole-body glucose and lipid metabolism through substrate utilization and endocrine signaling pathways [12]. Several studies have confirmed the association between low muscle mass and NAFLD, indicating that reduced muscle mass is an important contributor to the development of NAFLD [13–18]. However, current evidence is predominantly derived from cross-sectional designs [11, 19]. In particular, the effect of skeletal muscle mass on the resolution of existing MASLD remains poorly understood, and evidence regarding the relationship between changes in skeletal muscle mass over time and the development or resolution of MASLD is scarce. In addition, the assessment of low muscle mass has primarily relied on techniques such as DXA, but the high cost and lack of portability limit its suitability for large-scale screening, especially in resource-limited settings [20]. Deep-learning techniques have revolutionized medical image processing, enabling exceptional performance in tasks such as image recognition, segmentation and analysis [21–23]. The automation and high efficiency of these techniques overcome the key limitations of traditional manual or semi-automatic segmentation methods, thus facilitating large-scale body composition (BC) analysis across patient cohorts.

Moreover, MASLD is a multifactorial disease, both genetic and environmental factors influence its development and progression. Previous genome-wide association studies (GWAS) on MASLD have identified key genetic variants linked to hepatic fat accumulation and disease pathogenesis in Asian populations. Among these, PNPLA3 and SAMM50 have been the most extensively studied. These variants influence MASLD susceptibility by regulating hepatic lipid metabolism, including triglyceride remodeling, lipid droplet accumulation and mitochondrial function, thereby promoting hepatic fat accumulation and disease progression [24–26].

Therefore, this study aims to investigate the independent impact of muscle mass changes and their interaction with genetic factors in the progression of MASLD in a Chinese population cohort by applying a fully automated deep learning pipeline to extract BC data at L3 level, with integration of metabolic parameters from health examinations.

Materials and methods

Study Subjects

The study subjects were derived from the Health Omics Preventive Examination Cohort (HOPE Cohort) of the Health Management Center at the First Affiliated Hospital with Nanjing Medical University. This research was conducted in accordance with the Declaration of Helsinki (as revised in 2024). The research protocol was approved by the Ethics Committee of the First Affiliated Hospital with Nanjing Medical University (2019-SR-209), and written informed consent was obtained from all participants.

The inclusion criteria were as follows: (1) received annual follow-up; (2) underwent abdominal CT, liver ultrasonography and blood tests at both baseline and during the follow-up period; (3) underwent the standardized questionnaires under the trained interviewers; (4) at least one follow-up. The exclusion criteria were (1) individuals with missing basic demographic information and important covariate information; (2) excessive alcohol consumers (defined as ≥ 20 g/day for women or ≥ 30 g/day for men) [27]; (3) viral hepatitis or other autoimmune liver diseases; (4) a history of malignancy; (5) poor-quality CT images. The detailed enrollment and exclusion process is illustrated in Figure S1. Finally, a total of 4269 participants with abdominal CT scans were recruited from May 2019 to December 2024, including 3139 without MASLD (incidence cohort) and 1130 with MASLD (resolution cohort) at baseline. The median follow-up was 1.8 years (interquartile range 1.0 to 2.2 years). In this study, participants were categorized according to the presence of MASLD at baseline. Subjects without MASLD were enrolled in the incidence cohort to observe the incidence of MASLD during follow-up. Subjects with MASLD were assigned to the resolution cohort to observe the resolution of MASLD during follow-up.

Acquisition of Clinical Data and Definitions of MASLD

Demographic characteristics, lifestyle factors, disease history, physical examination and laboratory parameters were obtained from the Hospital Information System (HIS). Physical examination included standardized measurements of body mass index (BMI), waist circumference (WC), systolic blood pressure (SBP)and diastolic blood pressure (DBP). Venous blood samples were collected after a minimum 10-hour overnight fast. Plasma total cholesterol (TC), triglycerides (TG), fasting blood glucose (FBG), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), aspartate aminotransferase (AST), alanine aminotransferase (ALT), gamma-glutamyl transferase (GGT) and uric acid (UA) were measured using standard enzymatic assays on an AU5800 automated biochemical analyzer (Olympus Corporation, Tokyo, Japan). Hypertension was defined as blood pressure greater than or equal to 140/90 mmHg or use of antihypertensive medication. Diabetes was defined as FBG ≥ 7.0 mmol/L or having a history of diabetes.

According to a Multi-Society Delphi Consensus established by the American Association for the Study of Liver Diseases (AASLD), the European Association for the Study of the Liver (EASL) and the Latin American Association for the Study of the Liver (ALEH) [28], MASLD is defined as hepatic steatosis and one or more of the five cardiometabolic risk factors with no other causes of steatosis. The assessment of hepatic steatosis was based on abdominal ultrasound performed by experienced radiologists. The five cardiometabolic risk factors for adult Asians are as follows:

1. Body mass index (BMI) ≥ 23 kg/m² or waist circumference > 94 cm for men and > 80 cm for women;

2. FBG ≥ 5.6mmol/L (100 mg/dL) or 2-hour post-load glucose levels ≥ 7.8 mmol/L (≥ 140 mg/dL) or glycated hemoglobin ≥ 5.7% (39 mmol/L) or diagnosed type 2 diabetes or treatment for type 2 diabetes;

3. Blood pressure ≥ 130/85 mmHg or specific antihypertensive drug treatment;

4. Plasma TG ≥ 1.70 mmol/L (150 mg/dL) or lipid-lowering treatment;

5. Plasma HDL-C ≤ 1.0 mmol/L (40 mg/dL) for men and ≤ 1.3mmol/L (50 mg/dL) for women or lipid-lowering treatment.

MASLD resolution was defined as the absence of hepatic steatosis on follow-up abdominal ultrasonography among participants with baseline MASLD. Follow-up ultrasonography was performed using the same standardized protocol by experienced ultrasonographers blinded to baseline MASLD status.

CT Image acquisition and assessment of skeletal muscle area and quality

Studies have shown that measuring skeletal muscle area and attenuation at the third lumbar vertebra (L3) level via abdominal CT scans has emerged as an efficient surrogate method for assessing whole body skeletal muscle mass [29]. Abdominal CT were obtained during routine physical examination using multidetector CT scanners from Siemens (SOMATOM Definition AS+) and Philips (IQon Spectral CT). Images were reconstructed with a slice thickness of 5 mm using standard reconstruction kernels (I30f/B30f for Siemens and STD with iDose level 4 for Philips). BC analysis was performed using a previously developed deep learning segmentation model that is publicly available through GitHub [21, 22, 30]. This automated pipeline has demonstrated excellent segmentation performance for skeletal muscle at the L3 vertebral level, with a Dice similarity coefficient of 0.97 and an excellent correlation with manual measurements (Pearson’s r = 0.99); external validation of 564 randomly selected CT examinations showed a segmentation error rate of only 2.5%, supporting the robustness and generalizability of the algorithm [31, 32]. The pipeline automatically selects axial CT slices at the L3 vertebra level through a fully convolutional segmentation network. Then, the selected CT images are automatically segmented to generate boundaries of total abdominal muscle area (TAMA) [14], The TAMA included all the muscles seen in the images, including the psoas, paraspinal, transversus abdominis, rectus abdominis, quadratus lumborum and internal and external obliques. The evaluation of mean abdominal muscle attenuation (MAMA) can be achieved through the quantification of the mean CT value of the abdominal muscle at L3 level.

The surrogate measures of TAMA included TAMA normalized by the square of the height (TAMI), weight (TAMA/weight) and body mass index (TAMA/BMI). The annual growth rates of TAMA/weight (ΔTAMA/weight) and TAMA/BMI (ΔTAMA/BMI) were calculated using the following formula: Δ = [(X₁ - X₀) / X₀]/Δt × 100%, where X₀ and X₁ represent baseline and follow-up values, respectively, and Δt denotes the follow-up time (years).

Genotype‑stratified analysis with MASLD outcomes

Genotyping was performed using Illumina ASA and GSA chips on the Illumina iScan System following the manufacturer’s protocols. After standard quality control, variants were filtered based on minor allele frequency (≥ 0.005) and Hardy–Weinberg equilibrium (p ≥ 5 × 10⁻¹⁰). To account for population stratification, genetic principal components were derived by merging the study samples with the 1000 Genomes Project reference panel. All analyses were performed using GRCh37 (hg19) as the reference genome. Genetic variants included in the present study were selected on the basis of previous GWAS, which have consistently reported significant associations between these variants and MASLD [24, 26, 33, 34]. For each selected variant, the allele previously reported to be associated with increased MASLD risk was defined as the risk allele. Based on the risk allele status of the genetic variant, participants were categorized into risk allele carriers and non-carriers. Individuals carrying one or two risk alleles were classified as risk allele carriers, whereas those without risk alleles were classified as non-carriers. We initially evaluated the association between these MASLD-related genetic variants and MASLD status in the baseline population. Genetic risk was subsequently assessed for MASLD incidence and resolution, and further examined for interactions between genetic risk and the association of skeletal muscle changes with MASLD outcomes.

Statistical analysis

Continuous variables with normal and skewed distributions should be presented as mean ± standard deviation and median with interquartile range, respectively. Categorical variables are expressed as percentages (%). Continuous variables were analyzed with Student’s t-test or Mann-Whitney U test, and categorical variables were analyzed with chi-square test.

To accommodate the significant physiological difference in skeletal muscle mass observed between men and women, we conducted sex-stratified analyses. The associations between muscle mass and incident MASLD were initially examined using Cox proportional hazards models, with muscle mass analyzed both continuously and by tertiles. In the MASLD resolution cohort, we evaluated the impact of the ΔTAMA/weight on the resolution of MASLD using logistic regression analysis. Model assumptions were assessed using the Schoenfeld residuals for Cox proportional hazards models and Hosmer-Lemeshow test for logistic regression. The results are expressed as either hazard ratio (HR) or odds ratio (OR) with corresponding 95% confidence intervals (CIs). Follow-up time was incorporated as the time-scale variable in the analysis. The workflow of this study is shown in Fig. 1.

Fig. 1.

Overview of the study workflow. Green represents subcutaneous adipose tissue, red represents total abdominal muscle, and yellow represents visceral adipose tissue. CT, computed tomography; L3, the third lumbar vertebra; MASLD, metabolic dysfunction-associated steatotic liver disease; RCS, restricted cubic spline; TAMA, total abdominal muscle area; MAMA, mean abdominal muscle attenuation; TAMA/weight, TAMA adjusted by body weight; ΔTAMA/weight=[ (X₁- X₀) / X₀ ] / Δt × 100 ; X=TAMA/weight; Δt = follow up time(years)

Multivariate Cox proportional hazard analysis was used to examine the independent risks of the development of MASLD. Adjustments for variables were as follows: Model 1, unadjusted; Model 2, adjusted for age, alcohol consumption, smoking status, and exercise; Model 3, further adjusted for WC in addition to the covariates in Model 2; and Model 4, variables in Model 3 plus SBP, DBP, FBG, TG. The adjusted HR and its 95%CI were calculated for each model. Restricted cubic splines (RCS) of continuous TAMA/weight with knots at 10th, 50th and 90th percentiles were applied to graphically assess the potential non-linear association between TAMA/weight and development of MASLD with adjustment for all potential confounding factors.

Subsequently, sensitivity analyses were performed to evaluate the robustness of the findings by considering obesity status and follow-up duration, as well as by using TAMA/BMI as an alternative normalization approach to TAMA/weight. In addition, to explore the impact of increased muscle mass on MASLD resolution across different genetic backgrounds, we performed logistic regression analyses stratified the risk genotype, examining the association between ΔTAMA/weight and MASLD resolution within each subgroup after adjusting for age, sex, smoking status, alcohol consumption, exercise, WC, SBP, DBP, FBG, TG and genetic principal components (PCs). All statistical analyses were performed using R statistical software version 5.0 (R Project for Statistical Computing). p < 0.05 was considered statistically significant.

Results

Baseline characteristics of the study population according to sex

In the MASLD incidence cohort, 3139 individuals who initially did not have MASLD were followed for the development of incident MASLD until December 31, 2024. The baseline characteristics of the study population are described in Table 1. During a median follow-up of 1.8 years (interquartile range, 1.0 to 2.2), 303 (33.3%) and 213(9.6%) individuals who developed MASLD in men and women, respectively. The mean age was 35.37 ± 8.90 years in men and 35.75 ± 7.98 years in women. Compared without MASLD individuals, individuals with MASLD exhibited higher levels of BMI, SBP, DBP, TC, TG, LDL-C, FBG, ALT, GGT and UA, whereas HDL-C was lower regardless of sex (all p < 0.05). Additionally, the skeletal muscle mass differed markedly between participants with and without MASLD, regardless of sex.

Table 1.

Baseline characteristics of the study population according to the presence of incidence MASLD

Variables	Men					Women
	Overall (N = 911)	No MASLD (N = 608)	MASLD (N = 303)	p		Overall (N = 2228)	No MASLD (N = 2015)	MASLD (N = 213)	p
Age (years)	35.37 (8.90)	35.15 (8.76)	35.82 (9.18)	0.289		35.75 (7.98)	35.42 (7.92)	38.84 (7.91)	< 0.001
Smoking (%):	190 (20.86%)	118 (19.41%)	72 (23.76%)	0.151		12 (0.54%)	10 (0.50%)	2 (0.94%)	0.321
Drinking (%):	161 (17.67%)	102 (16.78%)	59 (19.47%)	0.361		45 (2.02%)	41 (2.03%)	4 (1.88%)	1
Regular exercise(%):	541 (59.39%)	362 (59.54%)	179 (59.08%)	0.95		759 (34.07%)	695 (34.49%)	64 (30.05%)	0.22
Hypertension(%):	88 (9.66%)	49 (8.06%)	39 (12.87%)	0.028		78 (3.50%)	56 (2.78%)	22 (10.33%)	< 0.001
Diabetes mellitus (%)	12 (1.32%)	6 (0.99%)	6 (1.98%)	0.228		4 (0.18%)	2 (0.10%)	2 (0.94%)	0.048
Metabolic parameters:
BMI( kg/m2)	23.37 (2.61)	22.70 (2.43)	24.72 (2.44)	< 0.001		21.38 (2.41)	21.10 (2.20)	23.99 (2.76)	< 0.001
WC (cm)	55.21 (12.01)	54.47 (14.04)	56.68 (5.93)	0.001		51.40 (10.85)	51.00 (11.08)	55.22 (7.32)	< 0.001
SBP(mm Hg)	121.92 (12.10)	121.19 (12.16)	123.38 (11.87)	0.01		113.48 (12.29)	112.86 (11.98)	119.31 (13.59)	< 0.001
DBP(mm Hg)	75.14 (9.01)	74.68 (8.84)	76.07 (9.27)	0.03		71.02 (8.78)	70.59 (8.60)	75.16 (9.40)	< 0.001
TC(mmol/L)	4.87 (0.88)	4.81 (0.85)	4.98 (0.94)	0.01		4.87 (0.89)	4.84 (0.87)	5.19 (0.98)	< 0.001
TG(mmol/L)	1.23 (0.90)	1.10 (0.52)	1.48 (1.35)	< 0.001		0.91 (0.41)	0.88 (0.35)	1.22 (0.70)	< 0.001
HDL-C(mmol/L)	1.31 (0.24)	1.34 (0.25)	1.24 (0.21)	< 0.001		1.53 (0.28)	1.54 (0.28)	1.42 (0.26)	< 0.001
LDL-C(mmol/L)	3.08 (0.66)	3.03 (0.65)	3.18 (0.68)	0.002		2.96 (0.64)	2.93 (0.63)	3.26 (0.72)	< 0.001
FBG(mmol/L)	4.97 (0.63)	4.93 (0.58)	5.04 (0.73)	0.028		4.87 (0.48)	4.85 (0.45)	5.03 (0.67)	< 0.001
UA((µmol/L)	378.44 (68.33)	371.94 (68.57)	391.49 (66.04)	< 0.001		273.47 (54.49)	271.08 (53.89)	296.13 (55.07)	< 0.001
NLR	1.73 (0.67)	1.72 (0.64)	1.77 (0.73)	0.339		1.85 (0.74)	1.83 (0.73)	2.06 (0.81)	< 0.001
ALT(U/L)	23.40 (13.07)	22.02 (12.52)	26.15 (13.73)	< 0.001		14.93 (12.74)	14.66 (12.70)	17.52 (12.87)	0.002
AST(U/L)	22.69 (8.53)	22.14 (6.40)	23.81 (11.62)	0.02		19.33 (9.57)	19.29 (9.64)	19.76 (8.93)	0.473
GGT(U/L)	29.12 (24.43)	26.93 (22.94)	33.51 (26.66)	< 0.001		17.14 (13.21)	16.73 (12.98)	21.10 (14.71)	< 0.001
BUN (mmol/L)	5.17 (1.15)	5.18 (1.12)	5.15 (1.20)	0.729		4.40 (1.03)	4.40 (1.04)	4.40 (0.97)	0.976
Creatinine (µmol/L)	81.36 (10.33)	81.42 (10.29)	81.22 (10.44)	0.784		58.46 (8.53)	58.54 (8.43)	57.68 (9.38)	0.198
Muscle measurements by CT:
TAMA(cm2)	153.10 (20.37)	150.82 (20.02)	157.68 (20.32)	< 0.001		99.28 (12.05)	98.64 (11.55)	105.33 (14.75)	< 0.001
MAMA(hu)	45.15 (4.17)	45.58 (4.28)	44.28 (3.80)	< 0.001		40.97 (4.75)	41.21 (4.69)	38.70 (4.80)	< 0.001
TAMI(cm2/m2)	50.53 (6.45)	49.81 (6.36)	52.00 (6.39)	< 0.001		37.95 (4.47)	37.66 (4.29)	40.76 (5.19)	< 0.001
TAMA/weight(cm2/kg)	2.17 (0.21)	2.20 (0.22)	2.11 (0.20)	< 0.001		1.78 (0.18)	1.79 (0.18)	1.71 (0.18)	< 0.001
TAMA/BMI	6.57 (0.70)	6.66 (0.71)	6.39 (0.65)	< 0.001		4.67 (0.53)	4.70 (0.53)	4.41 (0.53)	< 0.001

BMI, body mass index; WC, waist circumference ; SBP, systolic blood pressure; DBP, diastolic blood pressure; TC, total cholesterol; TG, triglyceride; HDL, high-density lipoprotein cholesterol; LDL-C, low-density lipoprotein cholesterol; FBG, fasting blood glucose; UA, uric acid; NLR, neutrophil-to-lymphocyte ratio; ALT, alanine aminotransferase; AST, aspartate aminotransferase; GGT, gamma-glutamyl transferase; BUN, blood urea nitrogen; MASLD, metabolic dysfunction-associated steatotic liver disease; TAMA, total abdominal muscle area; MAMA, mean abdominal muscle attenuation; TAMI, total abdominal muscle index; TAMA/weight, TAMA adjusted by body weight; TAMA/BMI, TAMA adjusted by BMI

In the MASLD resolution cohort, a total of 1130 individuals with MASLD at baseline were included, 162 (19.7%) and 77 (25.0%) patients experienced resolution of MASLD during the 4 year follow-up period in men and women, respectively. The mean age was 37.17 ± 8.62 years in men and 41.10 ± 10.12 years in women (Table S1). Compared with those who did not achieve MASLD resolution, individuals with MASLD resolution exhibited more favorable metabolic profiles, including lower BMI, lower TG, and lower UA levels, along with higher HDL-C, whereas no significant differences in blood pressure or FBG. Additionally, among men, those with MASLD resolution showed significantly lower levels of ALT and AST, while no such differences were observed in women.

Associations between low muscle mass and MASLD incidence

To evaluate the impact of skeletal muscle mass on MASLD development, we conducted multivariable-adjusted Cox regression analyses. We adopted the Schoenfeld residual method to assess the proportional hazard assumption and no violation was found. The TAMA and TAMI were positively associated with MASLD risk in both men and women, whereas TAMA/weight, TAMA/BMI and MAMA were inversely associated. Specifically, in men, per SD increase in TAMA (HR 1.28, 95% CI 1.15–1.43) and TAMI (HR 1.29, 95% CI 1.16–1.44) was associated with a 28% and 29% increased risk of MASLD, respectively. Conversely, per SD increase in TAMA/weight (HR 0.71, 95% CI 0.63–0.79), TAMA/BMI (HR 0.74, 95% CI 0.66–0.82) and MAMA (HR 0.79, 95% CI 0.71–0.88) was associated with a 29%, 26% and 21% decreased risk, respectively. In women, similar associations were observed: each one SD increase in TAMA (HR 1.74, 95% CI 1.52–1.98) and TAMI (HR 1.83, 95% CI 1.62–2.06) was associated with a 74% and 83% increased risk, respectively, and each one SD increase in TAMA/weight (HR 0.63, 95% CI 0.55–0.72), TAMA/BMI (HR 0.60, 95% CI 0.52–0.69)and MAMA (HR 0.66, 95% CI 0.58–0.74) was associated with a 37%, 40% and 34% decreased risk, respectively (Table 2). The results remained statistically significant after multivariable adjustment for potential confounders. Given that weight-adjusted muscle indices better account for body size and metabolic burden, TAMA/weight was selected as the primary measure, and TAMA/BMI was evaluated in additional analyses.

Table 2.

Sex-stratified HRs of MASLD according to skeletal muscle mass in the incidence cohort

Variables	Model 1			Model 2			Model 3			Model 4
	HR per SD (95%CI)	p		HR per SD (95%CI)	p		HR per SD (95%CI)	p		HR per SD (95%CI)	p
Men
TAMA (+ 20.37 cm 2 )	1.28 (1.15–1.43)	P < 0.001		1.32 (1.18–1.48)	P < 0.001		1.31 (1.17–1.47)	P < 0.001		1.29 (1.16–1.45)	P < 0.001
TAMI (+ 6.45 cm 2 /m 2 )	1.29 (1.16–1.44)	P < 0.001		1.32 (1.18–1.48)	P < 0.001		1.31 (1.17–1.46)	P < 0.001		1.29 (1.15–1.44)	P < 0.001
TAMA/weight (+ 0.21 cm 2 /kg)	0.71 (0.63–0.79)	P < 0.001		0.70 (0.62–0.79)	P < 0.001		0.71 (0.63–0.80)	P < 0.001		0.72 (0.63–0.81)	P < 0.001
TAMA/BMI	0.74 (0.66–0.82)	P < 0.001		0.71 (0.63–0.81)	P < 0.001		0.72 (0.64–0.82)	P < 0.001		0.74 (0.65–0.84)	P < 0.001
MAMA (+ 4.17 hu)	0.79 (0.71–0.88)	P < 0.001		0.76 (0.67–0.85)	P < 0.001		0.77 (0.68–0.87)	P < 0.001		0.79 (0.69–0.89)	P < 0.001
Women
TAMA (+ 12.05 cm 2 )	1.74 (1.52–1.98)	P < 0.001		1.78 (1.57–2.02)	P < 0.001		1.74 (1.54–1.98)	P < 0.001		1.62 (1.42–1.84)	P < 0.001
TAMI (+ 4.47 cm 2 /m 2 )	1.83 (1.62–2.06)	P < 0.001		1.84 (1.63–2.08)	P < 0.001		1.80 (1.60–2.03)	P < 0.001		1.73 (1.53–1.95)	P < 0.001
TAMA/weight (+ 0.18 cm 2 /kg)	0.63 (0.55–0.72)	P < 0.001		0.66 (0.58–0.76)	P < 0.001		0.66 (0.57–0.76)	P < 0.001		0.70 (0.60–0.81)	P < 0.001
TAMA/BMI	0.60 (0.52–0.69)	P < 0.001		0.63 (0.55–0.73)	P < 0.001		0.62 (0.54–0.72)	P < 0.001		0.64 (0.55–0.74)	P < 0.001
MAMA (+ 4.75 hu)	0.66 (0.58–0.74)	P < 0.001		0.69 (0.60–0.80)	P < 0.001		0.68 (0.59–0.78)	P < 0.001		0.71 (0.61–0.82)	P < 0.001

MASLD, metabolic dysfunction-associated steatotic liver disease; TAMA, total abdominal muscle area; MAMA, mean abdominal muscle attenuation; TAMI, total abdominal muscle index; TAMA/weight, TAMA adjusted by body weight; TAMA/BMI, TAMA adjusted by BMI; CI, confidence interval; HR per SD, hazard ratio per standard deviation increase; Model 1: unadjusted; Model 2: Adjusted for age, alcohol consumption, smoking status, and exercise; Model 3: Adjusted for age, alcohol consumption, smoking status, exercise, waist circumference; Model 4: Adjusted for age, alcohol consumption, smoking status, exercise, waist circumference, SBP, DBP, FBG, TG

As shown in Fig. 2, lower tertiles of TAMA/weight at baseline was significantly associated with higher risks of incident MASLD in both unadjusted and adjusted models regardless of sex. Compared with the highest tertile (Tertile 3) of TAMA/weight, the unadjusted HRs of the lowest tertiles (Tertile 1) were 2.52 (95% CI, 1.86 to 3.42) in men and 2.68 (95% CI, 1.86 to 3.86) in women; in Model 2, the HRs of the lowest tertile were 2.54 in men and 2.41 in women; after further adjustment for WC, the HRs of the lowest tertile were 2.49 in men and 2.42 in women; in Model 4, the HRs for the lowest tertile were 2.39 in men and 2.11 in women (all p < 0.001). RCS models with three knots were employed to assess the dose-response relationship between TAMA/weight and the prevalence of MASLD. In men, the analysis revealed a linear decrease in MASLD risk with increasing TAMA/weight levels (p = 0.093), and in women, a similar linear decreasing trend was observed (p = 0.729) (Figure S2). Additional analyses using TAMA/BMI as an alternative normalization method demonstrated findings consistent with the primary analyses based on TAMA/weight (Table S2).

Fig. 2.

Sex-stratified HRs of MASLD according to TAMA/weight tertile. The HRs of MASLD according to muscle mass tertile in men (A) and women (B). HR, hazard ratio; CI, confidence interval. Model 1: unadjusted; Model 2: Adjusted for age, alcohol consumption, smoking status, and exercise; Model 3: Adjusted for age, alcohol consumption, smoking status, exerciseand waist circumference; Model 4: Adjusted for age, alcohol consumption, smoking status, exercise, waist circumference, SBP, DBP, FBG and TG

Sensitivity analysis

Sensitivity analyses stratified by BMI (< 24 and ≥ 24 kg/m² for both men and women) and follow-up duration (< 2 and ≥ 2 years).In the analysis of TAMA/weight, each one SD increase was associated with a reduced risk of MASLD among men ( BMI < 24 kg/m²; HR = 0.77, 95% CI = 0.64–0.92, p < 0.05) and women (BMI < 24 kg/m²; HR = 0.73, 95% CI = 0.61–0.88, p < 0.05), with no significant interaction observed (p for interaction > 0.05 for both men and women). The analysis of TAMA/weight showed that each one‑SD increase significantly reduced the risk of MASLD for both sexes, regardless of follow‑up duration, with HRs of 0.68 (< 2 years) and 0.76 (≥ 2 years) for men, and 0.67 (< 2 years) and 0.64 (≥ 2 years) for women. The protective effect of muscle mass on MASLD did not significantly change with longer follow-up in either men or women, with no significant interaction observed in both sexes (all p for interaction > 0.05), suggesting that this protective effect remained stable throughout the follow-up period and was independent of follow-up duration. These associations remained consistent after multivariate adjustment (Table S4).

Associations between change in skeletal muscle mass and resolution of MASLD

Next, we further explored the association of baseline TAMA/weight and ΔTAMA/weight with the resolution of existing MASLD. To visualize the relationship of TAMA/weight, ΔTAMA/weightand MASLD resolution, we generated sex-stratified Sankey diagrams (Fig. 3). The diagrams illustrated among the 162 men with MASLD resolution, 56.2% were from the group with ΔTAMA/weight > 0, and 43.8% were from the group with ΔTAMA/weight ≤ 0. The results revealed that an increase in ΔTAMA/weight was associated with higher MASLD resolution rates. Additionally, 31.5% of the 162 men with MASLD resolution were from Tertile 1 and 34.6% were from Tertile 3, indicating that baseline muscle mass had little effect on resolution. For women, however, MASLD resolution was influenced not only by ΔTAMA/weight but also significantly by baseline TAMA/weight. Specifically, among the 77 women who achieved resolution, 53.2% belonged to the ΔTAMA/weight > 0 group, and 46.8% were from the ΔTAMA/weight ≤ 0 group. Moreover, of the individuals achieving resolution, 46.8% were from Tertile 3 and 23.4% were from Tertile 1. Higher baseline TAMA/weight combined with an increase in ΔTAMA/weight was associated with higher resolution rates.

Fig. 3.

Sankey diagram of the baseline TAMA/weight-ΔTAMA/weight-Resolution status in MASLD resolution cohort. Three columns from left to right indicating TAMA/weight, ΔTAMA/weight and resolution status respectively, the size of each rectangle representing the degree of connectivity; TAMA/weight, TAMA adjusted by body weight; ΔTAMA/weight=[ (X₁- X₀) / X₀ ] / Δt × 100 ; X=TAMA/weight; Δt = follow up time(years)

Cox proportional hazard models confirmed a positive relationship between baseline TAMA/weight and the resolution of MASLD in women (HR = 1.38, 95% CI 1.11–1.73, p = 0.004, Table 3A). Additionally, baseline TAMA was negatively associated with MASLD resolution in both sexes (HRs = 0.66 for men and 0.80 for women, Table 3A), consistent with findings from the MASLD incidence cohort.

Table 3.

Association of baseline TAMA, TAMA/weight and ΔTAMA/weight with resolution of MASLD

(A)
Variables	Model 1		Model 2		Model 3		Model 4
	HR per SD (95%CI)	p	HR per SD (95%CI)	p	HR per SD (95%CI)	p	HR per SD (95%CI)	p
Men
TAMA (+ 21.73 cm2)	0.66 (0.56–0.78)	< 0.001	0.66 (0.55–0.78)	< 0.001	0.71 (0.59–0.85)	< 0.001	0.70 (0.58–0.84)	< 0.001
TAMA/weight (+ 0.21 cm2/kg)	1.03 (0.88–1.20)	0.754	1.02 (0.87–1.19)	0.794	0.95 (0.81–1.12)	0.575	0.95 (0.81–1.13)	0.577
Women
TAMA (+ 15.69 cm2)	0.80 (0.63–1.01)	0.058	0.76 (0.59–0.98)	0.031	0.78 (0.60-1.00)	0.052	0.81 (0.62–1.06)	0.120
TAMA/weight (+ 0.18 cm2/kg)	1.38 (1.11–1.73)	0.004	1.38 (1.10–1.73)	0.005	1.36 (1.08–1.71)	0.009	1.38 (1.09–1.74)	0.008

(B)
Variables	Model 1		Model 2		Model 3		Model 4
	OR (95%CI)	p	OR (95%CI)	p	OR (95%CI)	p	OR (95%CI)	p
Men
ΔTAMA/weight ≤ 0	Reference		Reference		Reference		Reference
ΔTAMA/weight > 0	1.81 (1.28–2.56)	P < 0.001	1.86 (1.31–2.64)	P < 0.001	1.94 (1.37–2.78)	P < 0.001	1.94 (1.35–2.78)	P < 0.001
Women
ΔTAMA/weight ≤ 0	Reference		Reference		Reference		Reference
ΔTAMA/weight > 0	1.75 (1.04–2.96)	P = 0.034	1.71 (1.02–2.90)	P = 0.044	1.69 (1.00-2.87)	P = 0.050	1.66 (0.97–2.85)	P = 0.062

(A) The HRs for MASLD resolution according to baseline TAMA and TAMA/weight. (B) The ORs for MASLD resolution according to ΔTAMA/weight; TAMA, total abdominal muscle area; TAMA/weight, TAMA adjusted by body weight; ΔTAMA/weight=[ (X₁ - X₀) / X₀ ] / Δt × 100 ; X=TAMA/weight; Δt = follow up time(years); HR per SD, hazard ratio per standard deviation increase; OR, odds ratio; 95%CI, 95% confidence interval; Model 1: unadjusted; Model 2: Adjusted for age, alcohol consumption, smoking status, and exercise; Model 3: Adjusted for age, alcohol consumption, smoking status, exercise, and waist circumference; Model 4: Adjusted for age, alcohol consumption, smoking status, exercise, waist circumference, SBP, DBP, FBG, and TG

To assess the impact of the ΔTAMA/weight on the resolution of MASLD, we performed a logistic regression analysis (Table 3B). The Hosmer-Lemeshow goodness-of-fit tests showed adequate model fit (all p > 0.05). Before adjusting for covariates, compared with subjects who had ΔTAMA/weight ≤ 0, those with ΔTAMA/weight>0 were significantly associated with the resolution of MASLD in men and women, with ORs (95% CIs) of 1.81 (1.28–2.56) and 1.75 (1.04–2.96), respectively (all p < 0.05). Additional analyses using TAMA/BMI and ΔTAMA/BMI as alternative normalization methods demonstrated associations comparable to those observed in the primary analyses based on TAMA/weight and ΔTAMA/weight, respectively (Table S3).

Combined effects of muscle mass and genetic variants on MASLD resolution

Given the relatively short follow-up duration, the potential contribution of genetic susceptibility to MASLD incidence may not be fully captured in the incidence cohort. Therefore, subsequent genetic analyses focused on the resolution cohort to investigate whether genetic risk modified the association between changes in skeletal muscle mass and MASLD resolution among individuals with established MASLD. In the MASLD resolution cohort, after multivariable adjustment, the PNPLA3 was significantly associated with MASLD resolution (adjusted OR 0.75, p < 0.05, Table S5). Subsequently, we explored whether genetic effect could modify this increased muscle mass. Increase in ΔTAMA/weight was significantly associated with MASLD resolution. The OR was 3.54 (95% CI: 1.90–6.61) in the PNPLA3 CC (non-carriers) group and 2.44 (95% CI: 1.46–4.13) in the CG + GG (risk allele carriers) group. Although the effect estimates appeared to differ across genetic risk groups, the interaction between PNPLA3 genotype and muscle mass change was not statistically significant. (Table 4).

Table 4.

Combined effect of PNPLA3 genotypes and ΔTAMA/weight on MASLD resolution

SNP	Variables	Events/Total	Adjusted OR per SD (95%CI)	p	p for interaction
PNPLA3 rs738409					0.908
CC	ΔTAMA/weight>0	28/84	3.54 (1.90–6.61)	<0.001
CG + GG		50/180	2.44 (1.46–4.13)	<0.001
CC	ΔTAMA/weight ≤ 0	26/133	1.73 (0.95–3.16)	0.072
CG + GG		33/223	1.00 (Reference)

ΔTAMA/weight=[ (X₁ - X₀) / X₀ ] / Δt × 100; X=TAMA/weight; Δt = follow up time(years); OR per SD, odds ratio per standard deviation increase; 95%CI, 95% confidence interval. Analysis adjusted for age, sex, smoking status, alcohol consumption, exercise, waist circumference, SBP, DBP, FBG, TG and genetic principal component

Discussion

In this large-scale cohort study, we analyzed the association between the newly proposed concept of MASLD and muscle quantity and quality. We observed that muscle mass was significantly associated with the incidence and resolution of MASLD among men and women in our cohort. Specifically, lower muscle mass was associated with a higher risk of MASLD incidence, and an increased annual rate of muscle mass change was associated with a higher likelihood of MASLD resolution.

In the MASLD incidence cohort, the results showed that low muscle mass was strongly associated with the incidence of MASLD in both men and women, which is consistent with previous studies [11, 19, 35, 36]. Biologically, sarcopenia and MASLD share common pathways including insulin resistance, hormonal dysregulation, systemic inflammation, myostatin and adipokine secretion and nutritional deficiencies [10, 37, 38]. These established biological pathways support the plausibility of the results. The TAMA and TAMI was positively correlated with the presence of MASLD, while MAMA, TAMA/weight and TAMA/BMI were negatively correlated with MASLD incidence. These associations remained significant after adjusting for various confounding factors. Both TAMA and TAMI are positively linked to MASLD risk, possibly because individuals with greater total abdominal muscle area are more likely to exhibit abdominal obesity and higher intermuscular adipose tissue [14], which explains their increased MASLD risk. Furthermore, muscle attenuation directly reflects tissue quality, with lower attenuation indicating detrimental fat infiltration, and TAMA/weight better captures the clinically relevant imbalance between muscle and fat mass. Therefore, using TAMA/weight helps remove the confounding effect of obesity and more accurately reveals the independent protective role of muscle against MASLD [39]. Although TAMA/BMI is also a valid approach for assessing relative muscle mass by accounting for adiposity and body size, BMI is incorporated into the cardiometabolic framework of MASLD and may introduce overlap with metabolic risk assessment. Therefore, TAMA/BMI was evaluated as an additional analysis, and comparable results were observed. In the sensitivity analysis, the protective effect of muscle mass against MASLD was only significant in women with normal weight (BMI < 24). This may be because lean MASLD could be partly attributed to metabolic dysregulation associated with reduced muscle mass [40]. However, the occurrence of MASLD in overweight or obese females is primarily dominated by obesity, and the effect of muscle mass on its development is masked. Higher muscle mass reduced MASLD risk in both sexes, and this protective effect remained stable across the entire follow-up period, with no sex-specific time-muscle interaction observed (all p for interaction > 0.05), suggesting that the metabolic benefit of muscle mass was not dependent on follow-up duration. Protection diminished over time in men and increased in women, likely due to chronic inflammation impairing muscle metabolic quality in men and compensatory muscle anabolism after estrogen loss in women [34].

In the MASLD resolution cohort, this study introduced the annual growth rate of muscle mass, as changes over time can more accurately reflect alterations in metabolic health, which may be influenced by factors such as exercise, nutrition, or disease progression, rather than relying on a single baseline measurement [38, 41]. Gender differences of muscle mass for the risk of MASLD have been reported in previous studies [42–44], consistent with these studies [45], the analysis revealed a notable sex-specific pattern, with women showing that both TAMA/weight and ΔTAMA/weight significantly influenced MASLD resolution, while in men, only the ΔTAMA/weight had a significant effect. This divergence may be attributed to hormonal and physiological differences between sexes. Specifically, the no significant impact of baseline muscle mass on MASLD resolution in men may be explained by a diminishing benefit from their generally higher baseline muscle, along with a limited role of androgens in supporting metabolic improvement [42]. In contrast, the stronger association observed in women may reflect differences in metabolic profiles and hormonal status between sexes, although the underlying mechanisms remain unclear [46, 47].

Additionally, in this study, we observed a trend suggesting that the protective effect of increased muscle mass on MASLD resolution was attenuated in high risk genotype carriers, indicating a potential interaction between genetic risk and muscle mass. Risk variants in PNPLA3 are infamous for increased hepatic fat content and MASLD progression, by regulating lipid droplet triglyceride mobilization and associating with steatosis severity, liver fibrosis, and elevated ALT levels, respectively [15, 25]. No significant interaction effect between genetic variants and muscle mass on MASLD resolution was observed, which may be largely explained by the small sample size and short follow‑up period, considering that genetic variants may act via long‑term cumulative exposure and their effect may be less pronounced over a short timescale. Consequently, patients with risk alleles may require a greater increase in muscle mass to achieve MASLD resolution, but this result should be further explored in future studies.

The study has several strengths. Firstly, we used a fully automated deep learning pipeline to identify the L3 vertebral level and perform tissue segmentation on non‑contrast abdominal CT scans. The pipeline completes both slice selection and segmentation within seconds, which facilitates large-scale analyses of abdominal CT scans [48]. Secondly, this is the first bidirectional longitudinal cohort study in a Chinese health‑screening population to examine the muscle mass and MASLD relationship, which not only addresses disease risk but also clarifies the role of muscle mass gain in improving disease outcomes. Finally, we have evaluated not only the significance of increased muscle mass but also its effect on MASLD resolution across different genetic backgrounds, revealing that a greater increase in muscle mass is required to promote MASLD resolution in individuals with high genetic risk.

However, the present study also has potential limitations. First of all, we defined hepatic steatosis using the ultrasonography, not by liver biopsy, which is the gold standard tool of diagnosis for MASLD [49]. Secondly , although this was a longitudinal study, the median follow-up duration was relatively short, which may have limited the assessment of the long-term incidence and resolution of MASLD. Thirdly, due to the limited sample size and follow-up time, the relationship of genetic factors and muscle mass was not significant in the resolution cohort and the relatively short follow-up period precludes a definitive determination of the direction of causality, it is possible that subclinical MASLD have not yet been identified [37]. Additionally, the data were derived from a single center cohort, which may limit the generalizability of our findings to broader populations. These remain key areas for future research.

In summary, the results of this study indicate that maintaining muscle mass and implementing targeted strategies to prevent metabolic syndrome and obesity are crucial for preventing MASLD. The study revealed a significant association between muscle mass and MASLD in Chinese adults, indicating that the potential impact of skeletal muscle mass on MASLD should be fully considered in clinical prevention and treatment strategies. Improving skeletal muscle mass may confer significant benefits in the prevention and management of MASLD.

Conclusions

Low TAMA/weight is associated with an increased risk of MASLD incidence and persistence. These findings suggest that maintaining skeletal muscle mass may be clinically relevant for MASLD prevention and management.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

MASLD

Metabolic dysfunction-associated steatotic liver disease

NAFLD

Nonalcoholic fatty liver disease

CT

Computed tomography

L3

The third lumbar vertebra

BMI

Body mass index

WC

Waist circumference

SBP

Systolic blood pressure

DBP

Diastolic blood pressure

FBG

Fasting blood glucose

TC

Total cholesterol

TG

Triglyceride

HDL-C

High-density lipoprotein cholesterol

LDL-C

Low-density lipoprotein cholesterol

UA

Uric acid

ALT

Alanine aminotransferase

AST

Aspartate aminotransferase

GGT

Gamma-glutamyl transferase

TAMA

Total abdominal muscle area

MAMA

Mean abdominal muscle attenuation

TAMI

Total abdominal muscle index

TAMA/weight

TAMA adjusted by body weight

TAMA/BMI

TAMA adjusted by BMI

ΔTAMA/weight

Annual growth rates of TAMA/weight

ΔTAMA/BMI

Annual growth rates of TAMA/BMI

HR

Hazard ratio

CIs

Confidence intervals

OR

Odds ratio

RCS

Restricted cubic splines

BC

Body composition

GWAS

Genome-wide association studies

HOPE Cohort

Health Omics Preventive Examination Cohort

HIS

Hospital Information System

AASLD

The American Association for the Study of Liver Diseases

EASL

The European Association for the Study of the Liver

ALEH

The Latin American Association for the Study of the Liver

PCs

Genetic principal components

Author contributions

Conceptualization, Q.Z. and C.S.; methodology, Y.P.W. and J.L.; formal analysis, Y.P.W.; investigation, Y.X.Y. and H.W.X.; data curation, J.L.Z., X.Y.G., W.T.Y. and S.M.M.; writing—original draft preparation, Y.P.W.; writing—review and editing, J.L.; supervision, C.X.Y., W.G., J.L.; funding acquisition, Q.Z.; All authors have read and agreed to the published version of the manuscript.

Funding

This study was supported by grants from the Noncommunicable Chronic Diseases-National Science and Technology Major Project (grant number: 2024ZD0524403), Jiangsu Province Capability Improvement Project through Science, Technology and Education (grant number: ZDXK202248), Jiangsu Province Frontier Technology R&D Program (grant number: BF2025615) and the Specialized Diseases Clinical Research Fund of Jiangsu Province Hospital (grant number: DL202411).

Data availability

The datasets generated during the current study are not publicly available due to laboratory policies, but are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study was conducted in accordance with the Decla-ration of Helsinki (as revised in 2024). The research protocol was approved by the Ethics Committee of the First Affiliated Hospital of Nanjing Medical University (2019-SR-209). All participants provided written informed consent.

Competing interests

The authors declare no competing interests.