Artificial intelligence in metabolic dysfunction-associated steatotic liver disease: A short narrative review of diagnostic and therapeutic implications

Abstract

Metabolic dysfunction-associated steatotic liver disease is the most prevalent chronic liver disease worldwide and may progress from steatosis to metabolic dysfunction-associated steatohepatitis, fibrosis, cirrhosis, and hepatocellular carcinoma. Insulin resistance, obesity, mitochondrial dysfunction, inflammation, and impaired autophagy contribute to its pathogenesis. Early diagnosis, monitoring of disease progression, and assessment of therapeutic response remain important clinical challenges. Artificial intelligence (AI) and machine learning may improve disease prediction by integrating laboratory, metabolic, imaging, histological, and molecular data. Natural language processing and deep-learning approaches applied to radiological imaging and digital histopathology may support automated quantification of steatosis, inflammation, and fibrosis, while AI-assisted multi-omics analysis may identify molecular signatures relevant to disease stratification. Current evidence is strongest for retrospective case identification, imaging-based classification, and quantitative digital pathology; therapeutic selection and monitoring remain largely investigational. AI should therefore be viewed as a decision-support technology rather than a replacement for clinical judgment.

Key Words: Metabolic dysfunction-associated steatotic liver disease; Artificial intelligence; Machine learning; Liver fibrosis; Digital pathology; Precision hepatology

Core Tip: Artificial intelligence (AI) can integrate clinical records, radiological imaging, digital histology, and molecular data to support metabolic dysfunction-associated steatotic liver disease detection and risk stratification. Evidence is strongest for retrospective identification, imaging-based classification, and reproducible histological quantification. In contrast, AI-guided treatment selection, lifestyle personalization, and therapeutic monitoring remain investigational. This review distinguishes clinically proximate applications from exploratory and preclinical uses and identifies the validation, fairness, implementation, and outcome requirements for responsible clinical translation.

INTRODUCTION

Metabolic dysfunction-associated steatotic liver disease (MASLD) is the most prevalent chronic liver disease worldwide[1-4]. The current terms MASLD and metabolic dysfunction-associated steatohepatitis (MASH) are used throughout this review. Historical terms, including non-alcoholic fatty liver disease (NAFLD), non-alcoholic steatohepatitis, and metabolic dysfunction-associated fatty liver disease, are retained only when describing earlier studies, established score names, or published article titles[2].

The pathogenesis of MASLD spans simple hepatic steatosis, MASH, advanced fibrosis, cirrhosis, and hepatocellular carcinoma. MASLD is diagnosed through evidence of hepatic steatosis together with one or more cardiometabolic risk factors, including overweight or obesity, type 2 diabetes mellitus, and metabolic dysregulation[2,4]. Insulin resistance and steatosis may be complicated by lipotoxicity and inflammation driven by DNA damage, mitochondrial dysfunction, lysosomal injury, endoplasmic-reticulum stress, and inflammatory cytokine release. These processes activate Kupffer cells and hepatic stellate cells, thereby promoting fibrosis, cirrhosis, and potentially hepatocarcinogenesis[4,5]. Autophagy dysfunction may also contribute to MASLD pathogenesis[4,6,7], while genetic susceptibility, nutrition, and environmental exposures are relevant to pediatric MASLD[8].

Pharmacological interventions, including antihyperglycemic and lipid-lowering agents, are used to manage the metabolic comorbidities associated with MASLD[3]. Lifestyle interventions, including exercise and selected dietary strategies such as intermittent fasting, may improve insulin resistance, reduce inflammation, and influence autophagy-related pathways[4,5,9,10].

Several recent reviews have examined artificial intelligence (AI) in MASLD. Nagaraj Sharma et al[11] conducted a scoping review focused on MASLD detection, fibrosis assessment, and hepatocellular carcinoma risk, whereas Niazi et al[12] provided a broad narrative account of AI in MASLD diagnosis and management. The present review does not claim an unreviewed topic; its specific contribution is to connect diagnostic, prognostic, therapeutic, and implementation applications while distinguishing clinically proximate evidence from investigational and preclinical uses. It also compares quantitative findings with their validation limitations and frames AI as decision support rather than an autonomous substitute for clinical judgment.

MATERIALS AND METHODS

This short narrative review used a focused literature search covering January 2016 to August 3, 2026. The 10-year period was selected to capture the expansion of modern machine learning, deep learning, natural language processing (NLP), radiomics, and digital pathology in MASLD research. Earlier publications were considered when needed for nomenclature, foundational concepts, or relevant clinical guidance.

The literature was identified through PubMed, Google Scholar, the Directory of Open Access Journals, and ScienceDirect. Search concepts combined MASLD and MASH, including historical nomenclature, with “artificial intelligence”, “machine learning”, “deep learning”, “natural language processing”, “radiomics”, “digital pathology”, “multi-omics”, “diagnosis”, “fibrosis”, “prognosis”, “treatment”, “monitoring”, and “clinical trial”. Search syntax was adapted to each platform, and bibliographic details of studies emphasized in this review were checked against PubMed or publisher records.

Eligible sources were English-language peer-reviewed original studies, clinical trials, guidelines, and systematic, scoping, or narrative reviews that directly addressed AI-enabled diagnosis, risk prediction, treatment support, response assessment, or translational development in MASLD. Studies using historical terminology were eligible when the population corresponded to the contemporary MASLD/MASH spectrum. General AI, biosensor, or liver-disease articles were excluded when their relevance to MASLD was indirect and did not inform a clearly identified application. Preprints and other non-peer-reviewed sources were not used to support strong clinical conclusions.

Study selection and synthesis were narrative rather than systematic. No protocol was registered, and duplicate screening, formal risk-of-bias assessment, and meta-analysis were not undertaken. Priority was given to studies reporting a clear reference standard, quantitative performance data, independent or external validation, and clinically interpretable outcomes. Evidence was described as clinically proximate, investigational, or preclinical according to study design, validation setting, and distance from routine patient care. The summary tables present selected landmark studies and are not intended as an exhaustive catalogue.

AI IN DIAGNOSIS OF MASLD

NLP in early diagnosis of MASLD

AI applied to electronic health records (EHRs) may support earlier identification of MASLD by combining structured clinical variables with information extracted from free-text reports. Machine-learning models can analyze biochemical, metabolic, and demographic data, whereas NLP can identify steatosis, fibrosis, and other relevant features in radiology, pathology, and clinical documentation[13]. In a cohort of 38575 patients from the Mount Sinai BioMe program, which included participants of Hispanic/Latino, African, European, and other ancestries, NLP and EHR data were used to identify patients at risk of progression; 28.4% of a radiology-identified subgroup later developed cirrhosis after a median of 1057.3 days[14]. Another algorithm was developed using 620 records from the Research Patient Data Registry at Partners Health Care and then applied to an independent cohort of 8458 individuals with a high probability of historically defined NAFLD[15]. These studies support EHR-based case finding and risk stratification, but their retrospective designs do not establish improved patient outcomes.

NLP-based systems have reported positive predictive values of approximately 93%-99% for identifying steatosis from imaging reports and clinical documentation[13]. In a 6-month study of 957 patients, an AI-supported EHR approach identified 140 patients (14.6%) with a MASLD-related diagnostic code[16]. An NLP algorithm developed for pathology reports also extracted steatosis, lobular inflammation, hepatocellular ballooning, and fibrosis from approximately 1200 reports[17]. More recently, You et al[18] developed interpretable machine-learning models using routine non-invasive clinical and laboratory data, with extreme gradient boosting (XGBoost) achieving an area under the receiver operating characteristic curve (AUROC) of 0.8738 in the test set. These findings are promising for scalable detection, although prospective external validation and assessment of workflow benefit remain necessary.

AI in diagnosis of MASLD using image processing

Deep-learning methods have been applied to ultrasonography, computed tomography (CT), magnetic resonance imaging (MRI), and elastography in MASLD and related liver disease settings. Convolutional neural networks (CNNs) can quantify imaging features associated with hepatic fat and fibrosis, but reported performance should be interpreted according to the study population, reference standard, and validation method.

In ultrasonography, 13608 images from 3446 patients who underwent surgical resection, biopsy, or transient elastography were used to train a deep CNN to predict METAVIR fibrosis stage. The model achieved 85.7%-90.1% accuracy and performed favorably compared with experienced radiologists for cirrhosis detection[19]. AI-enhanced elastography may also improve tissue stiffness analysis and reduce observer variability through multilayer perceptrons, CNNs, and recurrent neural networks[20]. These results remain dependent on local acquisition protocols and require external testing before routine use.

Radiomics and physics-aware AI across CT and MRI can support automated liver segmentation and quantitative mapping of proton density fat fraction, fibrosis-related features, and iron deposition[21]. Transfer learning has also been applied to B-mode ultrasonography for non-invasive steatosis grading[22]. In a prospective multicenter study, 398 of 654 eligible patients contributed 1990 elastography images; deep-learning radiomics showed better overall fibrosis staging performance than two-dimensional shear-wave elastography images or biomarkers alone[23]. Because that study involved chronic hepatitis B, its applicability to MASLD requires direct validation. Imaging AI is clinically attractive because it can use routinely acquired data and reduce observer variation, but translation depends on scanner and protocol harmonization, transparent handling of image quality, external testing, and evidence that model output changes management.

AI-assisted digital pathology in diagnosis of MASLD

Histopathology remains important for assessing steatosis, inflammation, ballooning, and fibrosis, but conventional scoring is affected by sampling variability and interobserver disagreement. Digitized whole-slide liver-biopsy images can be analyzed with deep-learning methods, particularly CNNs, to quantify these features. Forlano et al developed a machine-learning system using 100 biopsy samples and validated it in an additional 146 samples; agreement for steatosis, inflammation, ballooning, and fibrosis was high[24].

AI-based quantitative pathology can generate continuous measurements for disease staging and treatment-response assessment, including fine architectural features that may be overlooked during routine visual review[25,26]. Recent assessments of AI-assisted MASH histology similarly identify potential gains in reproducibility while emphasizing unresolved issues in staining variation, scanner compatibility, annotation quality, endpoint definition, and regulatory acceptance[27]. Accordingly, AI-assisted pathology should be viewed as an adjunct to expert review and a candidate quantitative endpoint, not as an established replacement for pathologists.

Biomarkers and multi-omics diagnosis of MASLD integrated with AI

Integrating AI with metabolomics, proteomics, transcriptomics, lipidomics, and genomics may help identify circulating biomarkers and molecular subtypes across the MASLD spectrum. Transcriptomic profiling of 206 histologically characterized samples identified gene signatures related to endoplasmic reticulum stress, inflammation, steatohepatitis, and fibrosis, including AKR1B10, GDF15, and PDGFA[28].

Metabolomic studies have identified altered amino acid, bile acid, and lipid pathways associated with oxidative stress and disease severity[29]. In a multiethnic Asian cohort, serum lipidomic signatures discriminated MASLD, at-risk MASH, and advanced fibrosis; the at-risk MASH predictor showed lower performance on independent validation, illustrating the need for transportability testing[30]. Genomic analyses have also implicated variants such as MBOAT7 across the histological spectrum of MASLD[31].

AI-assisted multi-omics integration may support biomarker discovery, molecular phenotyping, and the identification of candidate therapeutic targets. However, standardization, model interpretability, validation across ancestries and care settings, and evidence of incremental clinical utility remain necessary before routine implementation.

Surface plasmon resonance in diagnosis of MASLD

Surface plasmon resonance is an emerging biosensing platform rather than a clinically validated AI application in MASLD. Reviews describe multiplexed, label-free sensing and machine-learning-assisted pattern recognition[32-35], while proof-of-concept sensor studies in other diseases illustrate technical feasibility but not MASLD-specific clinical utility[36]. Proposed MASLD applications include profiling hepatokines, adipokines, inflammatory mediators, extracellular vesicles, and fibrosis-related proteins. At present, these uses remain exploratory and require MASLD-specific assay standardization, prospective validation, and comparison with established non-invasive tests. The principal studies evaluating AI-assisted diagnosis in MASLD and related liver-disease settings are summarized in Table 1. Figure 1 illustrates the multimodal pathway linking clinical, laboratory, imaging, histological, and molecular data with AI-assisted diagnosis and prognostic assessment. Key studies evaluating AI-assisted prognostication and fibrosis-risk assessment in MASLD are summarized in Table 2.

Figure 1 Multimodal artificial intelligence pathway for metabolic dysfunction-associated steatotic liver disease diagnosis and prognosis. AI: Artificial intelligence; ALT: Alanine aminotransferase; AST: Aspartate aminotransferase; CT: Computed tomography; EHR: Electronic health record; MASLD: Metabolic dysfunction-associated steatotic liver disease; MRI-PDFF: Magnetic resonance imaging proton-density fat fraction.

Table 1 Artificial intelligence-assisted diagnosis in metabolic dysfunction-associated steatotic liver disease and related liver-disease settings.

Ref.	Data/model	Reference standard and validation	Main result	Focused limitation
Corey et al[15]	Derivation cohort of 620 patients; adaptive LASSO logistic-regression algorithm using EHR narrative mentions, ICD-9 codes, and triglyceride levels; subsequently applied to 8458 individuals	Manual chart review with biopsy or clinical diagnosis as the reference standard	The combined algorithm achieved an AUC of 0.85, compared with 0.75 for ICD-9 codes alone, and identified an independent cohort of 8458 individuals with a high probability of historically defined NAFLD	Retrospective design; historical terminology; effects on clinical outcomes were not evaluated
Stuart et al[16]	NLP-supported EHR identification of 957 patients with MASLD over six months	Manual review of monthly cohorts; NLP identification of steatosis from abdominal imaging reports and structured assessment of MASLD criteria	Overall PPV exceeded 93%; the imaging-report NLP component achieved a PPV of up to 99.4%; only 140 of 957 patients (14.6%) had a MASLD-related diagnosis code	Single-health-system study; dependent on documentation and coding practices; no external validation
Lee et al[19]	Deep convolutional neural network trained on 13608 ultrasonography images from 3446 patients; internal test set of 266 patients and external test set of 572 patients	Pathological specimens or METAVIR scores estimated by transient elastography; comparison with five radiologists	Four-class METAVIR accuracy was 83.5% internally and 76.4% externally. Cirrhosis AUC was 0.901 internally and 0.857 externally; external performance exceeded that of the five radiologists	Not MASLD-specific; reference standards were heterogeneous; tertiary-centre transportability requires evaluation
Wang et al[23]	Prospective multicentre study of 398 patients and 1990 shear-wave elastography images from 12 hospitals; deep-learning radiomics model	Histological fibrosis staging in patients with chronic hepatitis B	AUCs were 097 for F4, 0.98 for ≥ F3, and 0.85 for ≥ F2; overall performance exceeded conventional two-dimensional shear-wave elastography and biomarkers for most fibrosis thresholds	Indirect evidence for MASLD because the study population had chronic hepatitis B
Forlano et al[24]	Machine-learning analysis of 246 liver biopsies: 100 in the derivation cohort and 146 in the validation cohort	Independent pathologist assessment using NASH Clinical Research Network criteria and manual histological annotations	In the derivation cohort, ICCs were 097 for steatosis, 0.96 for inflammation, 0.94 for ballooning, and 0.92 for fibrosis. Quantitative findings were subsequently confirmed in the validation cohort	Single-centre biopsy cohort; sampling variability and transportability across staining and scanner platforms remain concerns
You et al[18]	Retrospective cohort of 152 patients; 17 routine clinical and laboratory variables; comparison of ten machine-learning models using five-fold cross-validation	MRI-PDFF-defined steatosis as the quantitative reference standard	Logistic regression achieved the highest AUC of 0.873. Random forest achieved the highest accuracy of 0.816 and F1 score of 0.807	Small, single-centre Taiwanese cohort; internal validation only; prospective external validation and assessment of clinical decision thresholds are required

AUC: Area under the curve; EHR: Electronic health record; ICC: Intraclass correlation coefficient; ICD-9: International Classification of Diseases, Ninth Revision; LASSO: Least absolute shrinkage and selection operator; MASLD: Metabolic dysfunction-associated steatotic liver disease; MRI-PDFF: Magnetic resonance imaging proton-density fat fraction; NAFLD: Non-alcoholic fatty liver disease; NASH: Non-alcoholic steatohepatitis; NLP: Natural language processing; PPV: Positive predictive value.

Table 2 Artificial intelligence-assisted prognostication and fibrosis-risk assessment in Metabolic dysfunction-associated steatotic liver disease.

Ref.	Population/data	Model/output	Main quantitative finding	Focused limitation
Van Vleck et al[14]	EHR data from 38575 participants; 2281 patients were identified with historically defined NAFLD, including 619 documented only in radiology reports	NLP combining structured and unstructured EHR data to identify disease and examine progression to NASH or cirrhosis	NLP achieved a sensitivity of 0.93 and an F2 score of 0.92. Among the 619 radiology-only cases, 170 patients (28.4%) were later considered likely to have developed NASH or cirrhosis after a median of 1057.3 days	Retrospective and documentation-dependent; historical terminology; progression was inferred from EHR documentation rather than prospectively adjudicated outcomes
Sherman et al[17]	More than 1200 pathology reports used for manual validation; prognostic cohort of 3134 patients with biopsy-confirmed MASLD followed for 20604 person-years	NLP extraction of steatosis, lobular inflammation, ballooning, and fibrosis stage from free-text pathology reports	Positive and negative predictive values ranged from 93.5% to 100%. Compared with simple steatosis, adjusted hazard ratios for cirrhosis were 1.04 for MASH/F0, 1.19 for MASH/F1, 1.89 for MASH/F2, and 4.21 for MASH/F3	Retrospective study within one multihospital health system; transportability across institutions and pathology-reporting practices requires evaluation
Njei et al[37]	Development cohort of 1487 NHANES 2017-2020 participants; external biopsy cohort of 337 patients; mortality cohort of 4276 NHANES III participants	Explainable XGBoost FibroX model for advanced fibrosis and long-term mortality prediction, interpreted using SHAP	FibroX achieved an AUROC of 0.97 vs 0.62 for FIB-4 in development and 0.84 vs 0.82 in the biopsy cohort. The mortality C-statistic was 0.80; the adjusted hazard ratio for cardiovascular mortality was 1.22	The development reference standard combined non-invasive assessment with transient elastography rather than biopsy; prospective clinical-impact and decision-threshold evaluation remain necessary
Njei et al[38]	Systematic review of 26 studies published during 2014-2025 across Europe, the United States, Asia, and South America	AI-based prediction of MASH and fibrosis stages ≥ F2-F4 using clinical, laboratory, imaging, and elastography variables	Validation/test AUROCs were approximately 0.76-0.95 for MASH, with external-validation performance as low as 0.67, and approximately 0.72-0.94 for fibrosis ≥ F2-F4	Considerable heterogeneity in populations, predictors, outcomes, and study designs; data diversity and model interpretability remain important barriers to clinical translation

AI: Artificial intelligence; AUROC: Area under the receiver operating characteristic curve; EHR: Electronic health record; FIB-4: Fibrosis-4 Index; MASH: Metabolic dysfunction-associated steatohepatitis; MASLD: Metabolic dysfunction-associated steatotic liver disease; NAFLD: Non-alcoholic fatty liver disease; NASH: Non-alcoholic steatohepatitis; NHANES: National Health and Nutrition Examination Survey; NLP: Natural language processing; SHAP: SHapley Additive exPlanations; XGBoost: Extreme gradient boosting.

AI IN THERAPY OF MASLD

Lifestyle intervention remains foundational in MASLD management, but current care also includes fibrosis risk stratification, longitudinal follow-up, and integrated management of obesity, type 2 diabetes, dyslipidemia, cardiovascular risk, and liver fibrosis[39,40]. Pharmacological options for selected patients with fibrotic MASH, including resmetirom and incretin-based therapies, are increasingly relevant to the therapeutic landscape[41-44]. In this setting, AI is best considered as a support tool for treatment selection, response monitoring, trial enrichment, and precision hepatology rather than as a replacement for clinician judgment[38,45-48].

The therapeutic relevance of AI in MASLD is related to disease heterogeneity. Patients with similar steatosis or fibrosis scores may differ in insulin resistance, adipose-tissue dysfunction, inflammatory activity, mitochondrial stress, genetic susceptibility, gut-liver interactions, cardiovascular risk, medication tolerance, and barriers to adherence[41,45,46,49]. Computational models may integrate clinical, biochemical, imaging, histological, behavioral, and molecular data to identify subgroups with different therapeutic needs[45,46,48]. However, direct evidence that such models improve treatment selection or patient outcomes remains limited. The main therapeutic applications of AI in MASLD are summarized in Figure 2 and Table 3.

Figure 2 Artificial intelligence-supported therapeutic pathway in metabolic dysfunction-associated steatotic liver disease. EHR: Electronic health record.

Table 3 Evidence relevant to artificial intelligence-supported therapy and therapeutic development in metabolic dysfunction-associated steatotic liver disease.

Ref.	Evidence base	Potential therapeutic contribution	Evidence maturity	Focused limitation
Naoumov et al[53]	Exploratory post hoc analysis of the FLIGHT-FXR trial involving 99 patients with F2-F3 NASH and 198 paired biopsies obtained before and after 48 weeks of placebo or tropifexor treatment	AI-assisted SHG/TPEF digital pathology quantified fibrosis, steatosis, ballooning, collagen architecture, and treatment-related histological changes. It identified reductions in overall and perisinusoidal fibrosis that were not detected by conventional ordinal scoring	Clinically proximate but investigational therapeutic-endpoint evidence	Post hoc analysis of a relatively small trial; evaluated treatment-response measurement rather than AI-guided treatment selection or improved patient outcomes
Harrison et al[42]	Phase 3 randomized trial involving 966 patients with biopsy-confirmed non-cirrhotic NASH and fibrosis who received resmetirom 80 mg, resmetirom 100 mg, or placebo	Demonstrates the expanding therapeutic context in which future decision-support systems may need to consider fibrosis stage, metabolic risk, treatment eligibility, safety, and monitoring requirements	Direct therapeutic evidence, but not an AI study	Does not validate an AI model or demonstrate that AI improves treatment selection, monitoring, or clinical outcomes
Muralidharan et al[30]	Serum lipidomic study of 151 patients with biopsy-paired MASLD data and independent validation in 105 patients; 481 serum lipids were evaluated	Identified lipidomic signatures that may inform future risk stratification, trial enrichment, or biomarker-based therapeutic research	Investigational biomarker evidence; not an AI treatment study	The study assessed disease classification rather than treatment response. The at-risk MASH model achieved an AUROC of 0.912 in development but declined to 0.76 during independent validation
Guo et al[49]	Review of inflammation-mediated crosstalk between diabetes and liver disease, including discussion of AI and systems biology	Provides mechanistic context for using computational approaches to characterize patient-specific inflammatory profiles and identify potential precision-therapy targets	Conceptual and hypothesis-generating evidence	No AI model was developed or validated, and no prospective AI-guided MASLD treatment comparison was conducted
Xia et al[55]	Preclinical drug-discovery study combining pose-filter ensemble learning with structure-based virtual screening of FXR agonists	Supported screening and lead identification, resulting in the discovery of XJ034, which showed FXR activity and metabolic effects in vitro and in high-fat-diet-induced obese mice	Preclinical proof-of-concept for AI-assisted drug discovery	No human MASLD evaluation, clinical treatment comparison, or patient-outcome evidence; the study supports lead discovery rather than established therapy selection
De Chiara et al[56]	Narrative review of the potential integration of AI with liver organ-on-chip platforms for historically defined NAFLD	Describes how AI could analyze complex organ-on-chip data to support target identification, drug screening, and preclinical efficacy or toxicity assessment	Conceptual and preclinical	The publication is a review and does not report direct experimental validation of an AI-integrated organ-on-chip platform in MASLD

AI: Artificial intelligence; AUROC: Area under the receiver operating characteristic curve; MASLD: Metabolic dysfunction-associated steatotic liver disease; MASH: Metabolic dysfunction-associated steatohepatitis; NAFLD: Non-alcoholic fatty liver disease; NASH: Non-alcoholic steatohepatitis; SHG/TPEF: Second-harmonic generation/two-photon excitation fluorescence.

AI-enabled risk stratification as a gateway to treatment selection

Risk stratification is an important potential application of AI in MASLD because treatment decisions depend on fibrosis stage, the likelihood of active MASH, cardiometabolic burden, and progression risk. Established tools such as Fibrosis-4 Index, the NAFLD fibrosis score, transaminases, and elastography are useful, but their accuracy varies across populations and is often less certain in intermediate-risk patients[38,39,47]. Machine-learning models can combine demographics, comorbidities, laboratory tests, medication history, imaging findings, and clinical records to estimate the probability of advanced fibrosis, active MASH, cirrhosis, hepatocellular carcinoma, and hepatic complications[45,47,50].

Reported AUROCs for machine-learning risk models range from approximately 0.72 to 0.95, but study populations, outcomes, and validation methods differ substantially[47,38]. FibroX, an explainable XGBoost model, reported an AUROC of 0.97 in development and 0.84 in an external biopsy cohort; its mortality C-statistic was 0.80[37]. Although encouraging, the reduction in performance across datasets illustrates the need for prospective evaluation at clinically relevant decision thresholds. The most credible near-term role is targeted support for confirmatory testing, specialist review, or closer follow-up; autonomous prognostic or treatment decisions are not justified by current evidence.

AI-assisted treatment selection in MASH

The availability of pharmacological therapies for selected patients with fibrotic MASH has increased the complexity of treatment decisions. Resmetirom, a thyroid hormone receptor-beta agonist, demonstrated histological benefit in a phase 3 trial involving patients with non-cirrhotic MASH and fibrosis[42]. Incretin-based therapies are also relevant to the broader metabolic treatment pathway[43,44]. These therapeutic advances do not, by themselves, validate an AI decision-support system. Fibrosis probability, steatohepatitis activity, body mass index, glycemic control, lipid profile, cardiovascular risk, renal function, contraindications, drug interactions, cost, access, and patient preference may all influence treatment planning[39,41,43,44].

In principle, an AI model could organize these variables to support referral or treatment discussions. The evidence is currently indirect: Most published models predict disease presence or fibrosis rather than compare treatment strategies. Any treatment-selection tool would require prospective validation, explicit decision thresholds, comparison with guideline-based care, and continued human oversight[45,46,48]. Claims that AI can optimize therapy should therefore be framed as a testable possibility rather than an established capability.

AI-driven personalized lifestyle intervention and digital phenotyping

Lifestyle intervention remains the therapeutic foundation of MASLD, yet durable behavior change is difficult to achieve[39,40]. AI-enabled mobile applications and wearable devices can record physical activity, dietary intake, sleep, body weight, glucose variability, blood pressure, heart rate, and medication adherence over time. These longitudinal data could support more individualized counseling, but MASLD-specific trials demonstrating sustained behavior change or improved liver outcomes remain scarce[45,46,48].

Digital phenotyping may identify early loss of engagement through declining activity, irregular sleep, reduced food logging, late-night eating, or worsening glucose variability. Such signals could prompt simpler goals, tailored reminders, behavioral coaching, or clinician review before relapse becomes clinically apparent[45,46]. This remains an investigational approach; guidance should account for patient readiness, domestic and workplace constraints, social barriers, and privacy considerations. Sustained weight loss may reduce steatosis and cardiometabolic risk, while greater weight loss is generally required to improve MASH and fibrosis[39,40].

Treatment response monitoring based on AI

Assessment of treatment response in MASLD should not rely on a single biomarker. Transaminases, body weight, glycemic control, lipids, platelet count, fibrosis markers, elastography, controlled attenuation parameter, MRI proton-density fat fraction (MRI-PDFF), and histology reflect different aspects of disease and may change independently. Normal transaminases do not exclude ongoing fibrosis risk, and elastography can be affected by inflammation, congestion, obesity, and measurement error[39,45,51]. AI could integrate longitudinal changes across these measures, but current evidence does not establish that it can determine treatment response without clinical interpretation[45,46,48].

Longitudinal integration may be more informative than a baseline risk score alone. Changes in alanine aminotransferase, aspartate aminotransferase, platelet count, haemoglobin A1c, lipids, body weight, liver stiffness, MRI-PDFF, medication exposure, and adherence could be reviewed together when deciding whether therapy should be continued, intensified, or reassessed. In patients with discordant results, AI may help flag the need for additional imaging, biopsy, specialist review, or surveillance[45,47,48]. Prospective studies are required to show that such support improves decisions or outcomes.

AI-assisted digital pathology as a therapeutic endpoint tool

Digital pathology is one of the more developed AI applications relevant to MASLD therapy, particularly in clinical trials. Histological endpoints remain important in MASH drug development, but biopsy is limited by sampling variability, uneven distribution of fibrosis and inflammation, and interobserver disagreement[27,39,51,52]. AI-assisted whole-slide analysis can quantify steatosis, ballooning, inflammation, collagen architecture, and fibrosis patterns with greater reproducibility than ordinal scoring alone, although standardized workflows and regulatory qualification are still required[51,52,27].

Quantitative pathology can generate continuous measurements of histological change. In a post hoc analysis of the FLIGHT-FXR trial, Naoumov et al[53] used second harmonic generation/two-photon excitation fluorescence microscopy with AI-based tools to assess fibrosis, steatosis, and ballooning after tropifexor treatment. The approach detected antifibrotic changes in patients with F3 fibrosis more sensitively than conventional microscopy. This is clinically relevant evidence for endpoint measurement, but it does not establish autonomous treatment selection. AI-assisted pathology remains an investigational adjunct to expert assessment[51-53].

AI-guided clinical trial enrichment and precision therapeutic development

MASLD and MASH trials are challenging because of high screen-failure rates, slow progression, placebo response, invasive endpoints, and biological heterogeneity. AI may help identify participants more likely to have active MASH, clinically significant fibrosis, measurable progression, or activation of a targetable pathway[38,45,46,48]. Combining non-invasive tests, imaging, clinical variables, and molecular features could reduce unnecessary biopsies and improve trial enrichment, but this proposed benefit requires prospective evaluation.

AI-based phenotyping may also support mechanism-specific trials. Therapies targeting thyroid hormone receptor beta, incretin signaling, fibroblast growth factor pathways, peroxisome proliferator-activated receptors, bile acid signaling, inflammation, mitochondrial dysfunction, or fibrogenesis are unlikely to benefit all patients equally[42-44,46,49]. Integrating metabolic, imaging, histological, and omics data could support hypothesis-generating treatment matching, but this application remains investigational.

AI in drug discovery, organ-on-chip platforms, and translational modeling

AI is increasingly used in early therapeutic discovery. MASLD involves overlapping mechanisms, including lipotoxicity, insulin resistance, oxidative stress and endoplasmic reticulum stress, mitochondrial dysfunction, innate immune activation, gut-derived inflammatory signaling, stellate cell activation, and extracellular matrix remodeling[41,45,46,49]. AI may help identify molecular networks, prioritize targets, predict compound-target interactions, support virtual screening, and explore drug repurposing opportunities[45,46,54].

Pose-filter-based ensemble learning has supported structure-based discovery of orally active nonsteroidal farnesoid X receptor agonists by accounting for receptor flexibility during virtual screening[55]. AI may also complement organ-on-chip platforms that model human liver microphysiology for preclinical drug testing[56]. Automated microfluidic systems supported by machine learning have been developed to reduce human error and improve high-throughput workflows[57]. These examples are preclinical or indirect and should not be interpreted as evidence of improved MASLD treatment outcomes[46,54,56].

AI-based multi-omics and combination therapy

Multi-omics integration may become relevant to MASLD therapy because patients with similar clinical features can differ in genomic, transcriptomic, proteomic, metabolomic, lipidomic, and microbiome profiles. AI is suited to analyzing these high-dimensional data and identifying subtypes associated with progression or treatment response[45,46,56]. Lipidomic signatures and inflammation-related molecular pathways may provide candidate features for future precision-treatment models[30,49], but prospective treatment-response validation is not yet established.

Combination therapy is likely to become more common as MASLD treatment options expand. Patients with MASH, obesity, dyslipidemia, type 2 diabetes, and cardiovascular risk may need several concurrent interventions. AI-based models could, in principle, compare combinations according to liver-related outcomes, cardiometabolic benefit, interactions, adverse effects, adherence, and patient preferences[45,46,48]. At present, this remains a research objective rather than a validated method for selecting therapy.

AI chatbots and clinician-supported patient counseling

Large language models and chatbots may have a limited but potentially useful role in MASLD care, including patient education, lifestyle counseling, medication explanation, clinic preparation, and routine workflow support. Studies suggest that their responses may be understandable, but accuracy is variable and expert oversight remains necessary[45,58,59]. Their use should therefore be restricted to supervised communication within validated clinical pathways.

These tools also carry risks. Hallucinated, outdated, incomplete, or overly confident information may mislead patients, particularly when medication use, fibrosis risk, or concerning symptoms are interpreted incorrectly. Large language models should not prescribe treatment, adjust medication, or replace clinician judgment[45,48,58]. Their safest current role is clinician-supervised education and workflow support with clear limits and escalation pathways.

VALIDATION, LIMITATIONS, AND ALGORITHMIC VIGILANCE

Models that affect referral, treatment initiation, monitoring intensity, or trial eligibility require more than strong discrimination. They should be calibrated at clinically relevant thresholds, useful for real decisions, fair across patient groups, and interpretable enough for clinicians to evaluate the output critically. Prospective testing in routine cohorts is therefore a minimum expectation[38,47,48].

Much of the current MASLD AI literature has not yet reached that level of evidence. Many studies are retrospective, use small or single-centre datasets, and are validated in settings close to the development population. Ethnic diversity, variation in metabolic phenotype, and differences in local testing pathways are not always represented, while calibration and decision-curve results are frequently missing[38,47]. These gaps matter clinically: Overestimated risk may lead to unnecessary testing or treatment discussions, whereas underestimated risk may delay referral or therapy.

Post-deployment vigilance should be part of therapeutic AI rather than an optional technical step. Clinic populations, coding practices, laboratory assays, imaging protocols, and available therapies change over time. A model that performs well during development can therefore lose accuracy when the data environment shifts. Performance monitoring, local revalidation, planned recalibration, and transparent updating are needed to protect safety and clinician confidence[47,48]. Equity requires similar attention. If ethnic, socioeconomic, or metabolic groups are underrepresented in training data, AI-guided pathways may reproduce disparities in access or outcomes rather than reduce them[47,48].

ETHICAL AND PRACTICAL IMPLEMENTATION OF THERAPEUTIC AI

Implementation of therapeutic AI requires resolution of several ethical and practical issues. Privacy is important because high-performing models may use EHRs, imaging data, pathology slides, wearable-device data, and behavioral information[45,48]. Explainability is also relevant because clinicians and patients may not accept recommendations when the reasoning is unclear[45,48]. Corporate influence, unequal access to technology, and automation bias may further affect how outputs are used. Ethical implementation should protect patient autonomy and confidentiality, preserve professional accountability, and strengthen rather than weaken patient-clinician interactions.

In practical terms, AI tools should be implemented without adding unnecessary bureaucracy. A risk score without a linked action pathway may have limited value. A well-designed model could help identify candidates for further assessment or treatment, suggest appropriate tests, prioritize follow-up, analyze historical trends, and support shared decision-making. The most appropriate model for MASLD is human-centered AI: Algorithms provide structured guidance while clinicians retain decision-making authority[45,47,48].

Collectively, AI technologies could contribute to earlier identification of patients who may benefit from treatment, more personalized lifestyle support, medication decision support, adaptive monitoring, trial enrichment, quantitative histological endpoints, and drug development. The greatest potential may lie in multimodal approaches combining clinical information, metabolomics, imaging, digital pathology, behavioral measures, longitudinal biomarkers, and molecular phenotyping[38,45-48]. However, AI should currently be viewed as a clinical decision-support tool rather than a self-sufficient means of determining treatment. Further development should prioritize prospective validation, comparative effectiveness, economic evaluation, fairness, patient-centered implementation, and demonstration of improved clinical outcomes.

CONCLUSION

MASLD is a major cause of chronic liver disease, and earlier diagnosis may help identify patients at risk of MASH, fibrosis, cirrhosis, and hepatocellular carcinoma. AI may support case identification, imaging-based classification, fibrosis-risk prediction, and reproducible histological quantification by integrating clinical, imaging, histological, and multi-omics data. The contribution of this short narrative review is to connect diagnostic, prognostic, therapeutic, and implementation applications while distinguishing clinically proximate evidence from investigational and preclinical uses. AI-guided treatment selection, personalized lifestyle support, and autonomous monitoring are not yet established. Translation into routine care will require prospective external validation, calibration, fairness assessment, workflow integration, cost-effectiveness evaluation, and evidence that AI-supported care improves patient outcomes. Until these requirements are met, AI should augment, not replace, multidisciplinary clinical judgment.