Advancing the diagnosis and treatment of metabolic-associated steatotic liver disease: The transformative role of artificial intelligence

Abstract

Metabolic-associated steatotic liver disease (MASLD), formerly referred to as non-alcoholic fatty liver disease, represents an escalating worldwide medical burden defined by hepatic steatosis, inflammation, fibrosis, and potential progression to cirrhosis or hepatocellular carcinoma. Artificial intelligence (AI) has emerged as a transformative tool in MASLD management, enhancing diagnostic accuracy, risk stratification, and treatment optimization. This review explores the integration of AI in MASLD diagnosis, including AI-based histopathological assessment, non-invasive screening models, imaging diagnostics, and gut microbiota-based approaches. Additionally, AI-driven treatment strategies facilitate personalized management, assess therapeutic response, and contribute to drug discovery. Despite its advantages, challenges such as data integration, model interpretability, and cost-effectiveness remain obstacles to widespread adoption. Future advancements in explainable AI, multi-modal data fusion, and cost-efficient implementations will be crucial for maximizing AI’s impact on MASLD care. AI-driven innovations hold great promise for improving early detection, guiding personalized treatment, and ultimately enhancing patient outcomes in MASLD.

Key Words: Metabolic-associated steatotic liver disease; Artificial intelligence; Non-invasive diagnosis; Liver fibrosis; Personalized treatment

Core Tip: Artificial intelligence (AI) is revolutionizing the diagnosis and treatment of metabolic-associated steatotic liver disease (MASLD) by enhancing diagnostic accuracy, enabling non-invasive assessments, and guiding personalized treatment strategies. Despite challenges such as data integration, model interpretability, and cost-effectiveness, AI has the potential to improve early detection, facilitate targeted interventions, and ultimately enhance patient outcomes in MASLD care.

INTRODUCTION

Metabolic-associated steatotic liver disease (MASLD), formerly designated as non-alcoholic fatty liver disease (NAFLD), constitutes a persistent hepatic disorder pathologically manifested through aberrant lipid deposition within hepatocytes, leading to inflammation, fibrosis, and the potential progression to cirrhosis or hepatocellular carcinoma (HCC). With the global rise in obesity, type 2 diabetes, and metabolic syndrome, the prevalence of MASLD has significantly increased, posing a considerable burden on healthcare systems worldwide. Traditional diagnostic and therapeutic approaches for MASLD, such as liver biopsy and histological assessment, have limitations including invasiveness, variability in interpretation, and subjectivity. When hepatic steatosis is less than 20%, ultrasound may miss it[1]. Recently, artificial intelligence (AI) has emerged as a revolutionary tool in the diagnosis, treatment, and management of MASLD, offering potential solutions to many of these challenges. This review explores the role of AI in enhancing MASLD diagnosis, guiding personalized treatments, and identifying novel therapeutic strategies, while discussing the challenges and future directions in AI implementation (Table 1).

Table 1 Artificial intelligence applications in metabolic-associated steatotic liver disease diagnosis and treatment.

Application	Description	Ref.
AI-based pathology tools	Improve histological scoring accuracy and reduce reader variability	Pulaski et al[3], Ratziu et al[4], Solomon et al[2]
Non-invasive screening models	Predict MASLD using lab and anthropometric data with high accuracy	Masaebi et al[5], Njei et al[6], Leow et al[10], Fan et al[7], Bosch et al[8]
AI-enhanced ultrasonography	Enhances steatosis and fibrosis diagnosis with improved AUC values	Santoro et al[11], Fujii et al[13], Luetkens et al[12]
Gut microbiota-based diagnosis	Identifies MASLD using ML models trained on gut microbiota data	Park et al[14]
Personalized treatment strategies	Guide targeted interventions using clinical and risk factor data	Dabbah et al[15], Wu et al[16], Malik et al[17]
Treatment response assessment	Evaluates therapeutic efficacy using continuous histological scores	Ratziu et al[4], Pulaski et al[3], Nishida et al[18]

AI: Artificial intelligence; AUC: Area under the curve; MASLD: Metabolic-associated steatotic liver disease; ML: Machine learning.

AI IN DIAGNOSIS OF MASLD

Enhanced histopathological assessment

AI-based pathology tools have made significant strides in improving the accuracy and reproducibility of histological evaluations in MASLD. Traditional liver biopsy is a gold standard for diagnosing MASLD; however, its invasive nature, risk of complications, and interobserver variability pose challenges. As highlighted by Solomon et al[2], the invasive nature of biopsy, coupled with risks of complications (e.g., bleeding and pain), sampling variability, and interobserver discrepancies, restricts its utility for large-scale clinical application. Furthermore, inconsistencies between non-invasive tests and biopsy results may stem from biopsy-related inaccuracies, such as sampling bias or staging subjectivity. AI models, such as AIM-MASH and AIM-NASH, have been developed to assist pathologists in scoring liver biopsies, improving the consistency and repeatability of assessments. These AI-driven tools have demonstrated high accuracy in scoring liver steatosis, inflammation, and ballooning while maintaining non-inferiority in fibrosis assessment[3]. Similarly, PathAI's NASH ML models offer continuous scoring for fibrosis, steatosis, and inflammation, providing a more quantitative approach to histopathological evaluation and increasing diagnostic precision[4].

Non-invasive screening and risk stratification

Non-invasive diagnostic tools have become a key focus of AI applications in MASLD, particularly for early detection and risk stratification. Machine learning (ML) algorithms have been developed to predict MASLD based on a combination of clinical, laboratory, and anthropometric data, including waist circumference, body mass index (BMI), alanine aminotransferase levels, and cholesterol levels. These AI models have demonstrated high accuracy [area under the curve (AUC) > 0.90] in identifying MASLD and stratifying patients based on their risk of developing advanced fibrosis or cirrhosis[5,6]. Fan et al[7] developed a logistic regression ML model [liver stiffness measurement (LSM)-plus] combining LSM with five clinical indices (age, sex, platelets, albumin, and total bilirubin). This model achieved a 96.8% accuracy for cirrhosis diagnosis in external validation, outperforming standalone LSM (86.4%) and Agile scores (76.0%), with a high specificity (97.4%) and sensitivity (88.9%). Bosch et al[8] devised a machine learning framework employing trichrome-stained hepatic biopsy specimens obtained from individuals diagnosed with cirrhosis secondary to MASLD. This computational tool ascertained pathologically relevant portal hypertension, achieving AUC values of 0.85 and 0.76 during the model training and external validation stages, respectively. Additionally, Lu et al[9] developed a random forest-based AI model integrating clinical features with the mitochondrial gene variant mt.12361A>G, which demonstrated exceptional diagnostic performance. The model achieved a perfect area under the receiver operating characteristic (AUROC) of 1.000 in the training cohort and maintained strong discriminative ability (AUROC = 0.876) in the validation cohort, significantly outperforming conventional statistical methods (P < 0.05). This imaging-free screening tool provides a practical solution for primary care settings. Furthermore, the LIVERSTAT test, an AI-powered diagnostic tool, uses blood biomarkers and anthropometric measurements to predict the risk of advanced fibrosis, demonstrating promising results in clinical studies[10].

AI-supported imaging diagnostics

AI-enhanced imaging techniques, particularly in ultrasonography, have revolutionized MASLD diagnosis. AI algorithms integrated into ultrasound systems have shown improved accuracy in detecting liver steatosis compared to manual assessments, with higher AUC values (0.87-0.98) and better sensitivity and specificity[11]. For instance, Luetkens et al[12] (2022) demonstrated that deep learning models (ResNet50/DenseNet121) could differentiate alcoholic from non-alcoholic cirrhosis on routine T2-weighted MRI with a 75% accuracy, outperforming traditional imaging assessments. AI-based analysis of liver contours and surface roughness has also been utilized to assess the severity of fibrosis, enabling non-invasive staging of MASLD[13]. These AI-driven imaging approaches provide a cost-effective, scalable alternative to liver biopsy, reducing patient burden and enhancing early detection.

Gut microbiota-based diagnosis

Emerging breakthroughs in gut microbial ecology have revealed that intestinal microbial communities exert a pivotal influence on the disease mechanisms underlying MASLD. AI models trained on 16S rRNA sequencing data have demonstrated high diagnostic accuracy (AUC > 0.90) in distinguishing MASLD from other liver diseases, such as alcoholic liver disease. These ML models use gut microbiota profiles as non-invasive biomarkers for MASLD, highlighting the potential of microbiome-based diagnostic tools in the clinical management of MASLD[14].

AI IN TREATMENT OF MASLD

Personalized treatment strategies

AI has paved the way for personalized medicine in MASLD by integrating clinical, imaging, and laboratory data to predict disease progression and guide individualized treatment strategies. Machine learning models can identify key risk factors, such as HbA1c, gamma-glutamyl transferase, and BMI, that influence disease progression and guide targeted interventions for high-risk patients[15,16]. By analyzing large datasets, AI can predict the likelihood of progression to advanced stages, including cirrhosis and HCC, enabling clinicians to implement early and tailored interventions.

Assessing treatment response

In the treatment of MASLD, evaluating therapeutic efficacy is crucial for optimizing care. Traditional methods, such as liver biopsy, may not always be suitable for monitoring disease response to treatment. AI-driven continuous scores have been used to assess changes in fibrosis, steatosis, and inflammation over time, offering a more sensitive and non-invasive method for evaluating treatment response. For example, clinical trials have demonstrated that AI-based scores were able to detect significant reductions in fibrosis and steatosis following therapeutic administration of semaglutide, a glucagon-like peptide-1 receptor agonist, which were not evident through conventional histopathological assessments[4].

Drug discovery and therapeutic insights

AI is also playing a vital role in the discovery of new therapeutic targets for MASLD. By analyzing large-scale clinical and molecular data, AI models can identify novel drug candidates and predict the effectiveness of potential treatments. This accelerates the drug discovery process and helps to prioritize compounds for further clinical development. AI-driven approaches are being used to predict treatment outcomes, providing valuable insights into the potential success of new therapies[17,18].

CHALLENGES AND FUTURE DIRECTIONS

Data integration and standardization

One of the primary challenges in implementing AI in MASLD is the integration of diverse data sources, including clinical, imaging, laboratory, and omics data. Current constraints encompass inconsistencies in algorithmic efficacy, absence of independent cohort verification, and reliance on meticulously curated data repositories. Harmonization of artificial intelligence algorithms and rigorous verification through multi-institutional investigations are imperative to guarantee their precision and clinical translatability. Additionally, establishing collaborative data ecosystems will prove indispensable for facilitating the broad integration of AI-driven diagnostic systems in healthcare settings[17,18].

Explainability and transparency

Machine learning systems, particularly those employing deep neural network architectures, frequently operate as "black boxes," impeding clinicians from understanding the decision-making process behind AI-driven diagnoses and treatment recommendations. To address this challenge, explainable AI techniques, such as SHAP analysis, are being developed to provide interpretable insights into the AI model’s predictions. Subsequent research priorities must strategically align artificial intelligence-driven systems with existing healthcare delivery frameworks, while establishing transparent algorithmic reasoning mechanisms to optimize therapeutic decision-making efficacy. Enhancing the transparency of AI algorithms will be crucial in building clinician trust and ensuring widespread adoption in clinical settings[6,16,19].

Cost-effectiveness and accessibility

For AI tools to be widely adopted, they must be both cost-effective and accessible, particularly in resource-limited settings. Certain computational frameworks, particularly those dependent upon high-dimensional omics datasets or sophisticated radiological modalities, encounter substantial deployment challenges in under-resourced healthcare environments[17]. The development of lightweight AI models that utilize readily available clinical data, such as liver function tests and anthropometric measurements, can facilitate the scalability and broader implementation of AI-based solutions. Ensuring that AI tools are affordable and can be easily integrated into existing healthcare infrastructure will be critical to their success[5,15].

Data privacy and security

The integration of AI in MASLD care relies on vast amounts of sensitive patient data, including clinical records, imaging studies, and multi-omics profiles. Preserving patient privacy and data security requires robust encryption and anonymization technologies, complemented by responsible data management practices such as decentralized sharing frameworks[20].

Regulatory frameworks and standardization

Divergent global regulatory landscapes pose barriers to AI adoption in health care. Whereas governing authorities including the United States Food and Drug Administration and European Medicines Agency have promulgated regulatory frameworks for artificial intelligence-based healthcare technologies, the absence of harmonized criteria to assess algorithmic reliability, therapeutic effectiveness, and sociodemographic parity persists as a critical implementation barrier[17]. This necessitates international collaboration to establish harmonized regulatory frameworks while resolving critical challenges including cross-jurisdictional liability and intellectual property standardization.

Ethical implications and algorithmic bias mitigation

AI applications in MASLD raise ethical concerns regarding fairness, accountability, and patient autonomy. Inherent biases in training datasets may disproportionately affect underrepresented populations, leading to disparities in diagnostic accuracy or treatment recommendations. Algorithmic bias presents a substantial impediment, necessitating heterogeneous data repositories and persistent oversight mechanisms to safeguard equitable outcomes[20]. Mitigation strategies include diversifying training cohorts, implementing fairness audits, and incorporating bias-correction algorithms. Through a strategic emphasis on patient-centric well-being, operational accountability, sociodemographic equity, and interdisciplinary coordination, stakeholders spanning medical systems, computational engineers, regulatory architects, and clinical practitioners can effectively harness artificial intelligence and autonomous systems to augment therapeutic paradigms, all while maintaining rigorous adherence to bioethical principles[20].

CONCLUSION

AI has significantly enhanced the management of MASLD by improving diagnostic accuracy, enabling non-invasive assessments, and facilitating personalized treatment strategies. While challenges such as data integration, model interpretability, and cost-effectiveness persist, ongoing advancements in AI promise to further transform MASLD care. AI-driven innovations offer the potential for earlier detection, more effective treatments, and ultimately improved patient outcomes. As AI technology continues to evolve, its integration into clinical practice will become increasingly essential for optimizing the care of patients with MASLD. AI is poised to revolutionize MASLD management in the next decade via multimodal monitoring, AI-driven pathology subtyping, and digital twin-validated personalized therapeutics.