Over the past two decades, hepatology practice has witnessed significant advancements in diagnostic, prognostic, and therapeutic fields. The integration of artificial intelligence (AI) into hepatology holds the potential to further transform the evaluation of complex clinical data, with AI applications demonstrating capabilities that, in specific tasks, surpass those of physicians (1, 2). In hepatology, AI has been utilized in areas such as liver imaging, histopathology interpretation, non-invasive testing, and evidence-based decision-making (1, 3, 4). It also enhances the analysis of large datasets, facilitating the advancement of precision medicine (5), and improving the efficiency and cost-effectiveness of telemedicine while supporting personalized, evidence-based care (6).

Specifically, AI applications have shown promise in identifying liver fibrosis, distinguishing different types of liver lesions, forecasting outcomes of chronic liver conditions, and suggesting management of metabolic- dysfunction steatotic liver disease (MASLD) (7). AI technologies have also been investigated for their potential to improve the diagnosis, prognosis, and treatment of hepatocellular carcinoma (HCC), a major challenge in hepatology (8).

Despite these advancements, the implementation of AI in hepatology faces several challenges, including data heterogeneity, ethical considerations, system interoperability, and the necessity for inclusive datasets that represent diverse populations (4). Issues related to data collection, standardization, and interpretation further hinder the widespread adoption of AI tools. Algorithms that exhibit deficiencies due to concerns related to data privacy and quality may pose potential risks when applied in clinical settings (3).

In the Middle East and North Africa (MENA) region, the adoption of AI in hepatology and gastroenterology has been relatively slow (9). This delay may be attributed to a lack of awareness, technical capacity, regulatory guidelines, and concerns regarding data privacy and clinical reliability. Understanding the perspectives of hepatologists on AI is crucial for addressing these challenges and effectively integrating AI-generated tools into clinical practice. Accordingly, we assessed hepatologists’ familiarity with AI, usage patterns, and concerns about its integration into practice to inform strategies for adoption across the MENA region. This work addresses a specific gap not covered by prior surveys in other regions and specialties (9–15): it is the first multinational assessment focused specifically on hepatologists from 17 MENA countries; it evaluates clinical and research use of AI alongside institutional readiness; and it integrates quantitative findings with a thematic synthesis of policy-level recommendations.

We received 236 completed responses that met inclusion criteria. Based on tracked seed and first-wave invitations (n = 285) disseminated via SLMENA channels, the estimated participation rate was 82.8%; because downstream snowball sharing was not fully traceable, a conventional response rate cannot be calculated. Participants were predominantly male (65.3%). The age distribution was balanced, with the largest groups being those aged 36–45 years (31.4%), followed by ages 25–35 (28.0%) and 46–55 years (23.7%).

The most common specialties reported were hepatology and gastroenterology, each accounting for 33.9% of participants. The respondents represented 17 MENA countries, primarily Turkiye (27.1%), Egypt (24.6%), and Oman (15.3%). Participants largely worked in their countries of origin, with minimal exceptions. Most respondents worked at academic or teaching hospitals (66.1%), followed by public hospitals (26.7%), with a significant majority serving as clinicians (91.9%).

Research activity among participants was moderate, with over 60% reporting between 1 and 5 peer-reviewed hepatology publications, and 20.3% having 6–25 publications. Professional experience was diverse, with 26.3% reporting over 20 years of practice and 28.0% having less than 5 years. Despite growing interest in AI, formal education in AI was notably limited; only 7.6% had completed formal AI training, and 45.8% had no prior AI education or exposure (Table 1; Figure 1).

To the best of our knowledge, this is the first study to comprehensively assess the adoption, perceptions, and challenges of AI in hepatology across the MENA region. Our findings reveal a notable gap between enthusiasm for AI and its practical implementation. Although AI is widely recognized for its potential to transform liver disease management, particularly in diagnostics, prognostics, and decision support, its integration into routine hepatology practice remains limited. This disconnect appears largely driven by deficits in specialized training, limited infrastructural capacity, and unresolved ethical and regulatory concerns. Our sample was concentrated in Türkiye, Egypt, and Oman—consistent with the SLMENA network’s reach and the snowball design—so estimates of AI awareness, routine use, and institutional readiness likely reflect these higher-responding settings. Countries with smaller samples may differ in material ways (e.g., data-sharing rules, EHR penetration, reimbursement models, workforce training), and results should therefore be generalized to the broader MENA region with caution. Descriptively, larger-sample countries reported greater exposure to AI tools and higher institutional willingness, plausibly tracking more mature digital infrastructure and governance. Across experience strata, enthusiasm for AI’s clinical potential was broadly shared, while formal AI training remained limited in all groups. Because the study was not designed for inference, these patterns are non-inferential and should be interpreted cautiously; future work should use stratified sampling frames with country-level quotas to enhance representativeness.

Overall, most participants in our survey expressed cautious optimism about the role of AI in hepatology, particularly its potential to improve diagnosis, early disease detection, and risk assessment. While secondary applications such as administrative support and treatment planning were rated lower, they were still viewed as valuable adjuncts to clinical workflows. Several respondents anticipated improved patient outcomes and expanded access to care, although concerns persisted regarding the reliability, cost, and potential to exacerbate existing disparities in access to technology. Trust in AI remained moderate, with a strong preference for human oversight, especially in patient-facing roles. Many participants expressed hope that AI would soon become a routine part of clinical practice. These findings align with previous reports documenting a shift in physicians’ attitudes, from initial fear of human replacement to a more measured optimism regarding AI integration, albeit tempered by concerns about diminished clinical autonomy (10, 11). Notably, such concerns appear to be more strongly influenced by user characteristics than by AI performance itself (10). Bisdas et al. noted that positive attitudes were more common among individuals with prior AI exposure or those practicing in well-resourced environments, while skepticism was more prevalent among medical students and respondents from low- and middle-income countries (11). This underscores the role of training and infrastructure in shaping perceptions. Despite persistent concerns related to control, ethics, and employment, more than 60% of physicians and trainees in other studies have expressed optimism about AI (12, 13), a sentiment shared by both gastroenterologists and general practitioners (14, 15). Compared with surveys of physicians/trainees in other regions and specialties (10–15), MENA hepatologists reported similar overall optimism but lower routine use, a stronger prioritization of imaging and fibrosis staging, markedly limited formal AI training, and a pronounced preference for human oversight—patterns that are consistent with infrastructural and governance gaps described in resource-constrained settings.

Our results show that only a small proportion of participants had received formal training in AI, with nearly half reporting no prior exposure. Despite this, there was strong interest in pursuing AI education, particularly when linked to their clinical practice or research work. Respondents favored hands-on workshops and simulation-based learning as preferred formats, and frequently emphasized the importance of integrating AI into medical education and fostering collaboration with AI experts. These findings underscore the urgent need for accessible, context-specific training in AI for hepatologists.

Structured AI education for hepatology professionals remains underdeveloped, consistent with trends across other medical fields. Studies have shown that most physicians and medical trainees lack formal AI instruction despite increasing integration of AI technologies into healthcare settings. For example, a survey from United Kingdom found that 92% of trainee doctors considered their current curricula inadequate for AI training (16). Likewise, a study from Nepal highlighted limited AI knowledge among medical students and interns (17). Across multiple studies, physicians have expressed a strong desire to improve their AI literacy and called for healthcare institutions and academic bodies to implement formal training programs (10, 14, 18). In line with these observations, structured, practice-oriented curricula and modular workshops can correct misconceptions and accelerate safe adoption (19).

Previous studies have shown that medical students often rely on media and peer discussions to learn about AI, due to limited formal training in their academic curricula (20, 21). In contrast, several structured educational programs have emerged in countries such as the United States (12, 13), Canada (221), France (22), and Mexico (23). These initiatives span a variety of instructional methods, including didactic lectures (22–26), discussion-based sessions (25), web-based modules (23), workshops, case-based formats (27), and experiential learning opportunities (24). Most of these programs are implemented in academic institutions (22–26), reflecting a growing institutional commitment to AI education.

The duration and design of these programs vary considerably, ranging from single-day workshops (25) to fellowships extending beyond 1 year (24), highlighting their adaptability to diverse learner needs. This flexibility enables educators to tailor content to different time constraints, experience levels, and learning goals. Such structured approaches have been shown to correct misconceptions, increase acceptance of AI, and reinforce its role as a supportive tool rather than a threat to clinical practice (11, 28, 29).

Integrating AI into healthcare education remains challenging. Institutional barriers such as conventional teaching methods, inflexible funding models, and restrictive university policies hinder innovation. Meaningful progress will require reforms in accreditation and licensing processes to create space for dynamic, forward-thinking curricula (28, 30). Inclusion of AI competencies into medical education must also be accompanied by a multidisciplinary approach that preserves core humanistic values, particularly compassion, which remain central to high-quality patient care (31).

Despite recent advances demonstrating the utility of AI in hepatology (7), our survey indicates that its adoption in the MENA region remains limited. Approximately two-thirds of respondents believed AI’s most significant value lies in imaging analysis and liver fibrosis staging. These views are supported by emerging evidence highlighting AI’s expanding role in managing MASLD. In particular, AI-driven predictive models have been developed to estimate HCC risk in MASLD populations (32). Beyond prediction, AI is increasingly used to enhance patient stratification, discover novel biomarkers, and identify therapeutic targets by analyzing data from electronic health records, digital pathology, and imaging (33).

AI has also shown promise in diagnostic support, especially in evaluating multimodal data such as imaging and laboratory findings (34). In the context of HCC and cholangiocarcinoma, deep learning models have been applied to improve diagnostic accuracy, tumor classification, and treatment planning, as well as to predict clinical outcomes (35–37). Similarly, in liver transplantation, AI is being utilized for dynamic risk prediction, optimizing organ allocation, and forecasting post-transplant outcomes (5). Despite these advancements, our data show that AI use in transplantation among hepatologists in the MENA region remains underdeveloped. Outside the MENA region, AI has already influenced hepatology through deep-learning image analysis for lesion characterization and staging, AI-assisted elastography/fibrosis grading, and multimodal decision support that integrates imaging with laboratory and clinical data (1, 3, 5, 7, 35–37). In parallel, clinical decision support systems increasingly target risk stratification and longitudinal care pathways, aligning with domain needs in hepatology where high-dimensional data and evolving phenotypes challenge traditional tools (1, 3, 5, 7).

In the research domain, just over 40% of our respondents reported using AI tools for tasks such as literature summarization, statistical analysis, and manuscript preparation. Researchers are increasingly recognizing both the potential of AI to advance research and the challenges it presents, particularly concerns related to data bias, lack of validation, and ethical considerations (1–3). Machine learning applications now support drug discovery (38), trial optimization (39), real-time data acquisition through wearable devices (40), and improved endpoint detection and risk monitoring (41). AI also plays a role in managing missing data and enhancing participant recruitment and retention using natural language processing and passive data collection techniques (42). However, persistent challenges, including data interoperability and secure data sharing, continue to hinder the broader application of AI in clinical research.

Participants noted limited institution-wide awareness of AI across the MENA region, highlighting a persistent gap between conceptual understanding and practical readiness for implementation. While many institutions demonstrated moderate preparedness, few appeared fully committed to integrating AI into their workflows. Although AI technologies have made substantial progress in healthcare, institutional readiness to adopt these innovations remains highly variable, hindered by multiple challenges and unresolved operational, infrastructural, and regulatory considerations (43). Respondents identified several key barriers to AI adoption in hepatology across the MENA region, including limited expertise, restricted access to AI tools, inadequate infrastructure, and unresolved regulatory issues. Ethical concerns were also prevalent, particularly regarding data security, algorithmic bias, and potential over-reliance on AI systems. These concerns align with findings from the broader international literature, which frequently cites the absence of ethical and legal standards as major impediments to AI integration in healthcare (15, 44, 45). To address these challenges, participants emphasized the need for clear regulatory frameworks, targeted training programs, rigorous clinical validation, and transparent collaboration between stakeholders. Data security and privacy risks have also been widely acknowledged (46). In addition, the lack of explainability, often referred to as the “black box” nature of AI, remains a challenge, as clinicians may struggle to interpret or trust AI-generated outputs (47). Another critical issue is the underrepresentation of minority and rural populations in AI training datasets, which undermines the generalizability and equity of AI models (48). Allocation of ethical and legal responsibility for AI-supported decisions remains jurisdiction-dependent and evolving. Existing guidance emphasizes that developers are accountable for design quality, data provenance, transparency, and post-deployment monitoring, whereas implementers and institutions are responsible for validation in-context, governance, and safe integration into clinical workflows; clinicians retain ultimate accountability for patient-facing decisions (4, 6, 48, 49). In the MENA region—where regulatory maturity and data-sharing rules vary—pragmatic “human-in-the-loop” deployment, institutional oversight, and clear audit trails are likely to be essential transitional safeguards until harmonized medico-legal frameworks are established. The concerns voiced by MENA hepatologists—privacy/confidentiality, bias and reliability, explainability, liability, and the necessity of human oversight—mirror priorities in international guidance (e.g., WHO recommendations) and EU-level initiatives emphasizing transparency, accountability, safety, and context-specific validation (4, 48, 50). Our results therefore reinforce the importance of adapting these global principles to local legal environments, data-governance capacity, and workflow realities.

In line with surveys from North America, Europe, and Asia, clinicians in our cohort reported high interest but tempered trust, with persistent concerns about privacy, liability, explainability, and cost (10–15, 18, 20, 44–47). Common barriers mirrored international reports, limited formal training, lack of integration with EHR/EMR systems, and unclear governance (10–15, 18, 20, 44–47). Relative to many high-income settings, however, MENA hepatologists reported lower routine use, more limited formal AI training, and stronger emphasis on infrastructure and regulatory readiness, consistent with challenges described in LMIC contexts (49). Notably, our respondents prioritized imaging and fibrosis staging more than some non-hepatology specialties, aligning with domain-specific opportunities in hepatology (1, 3, 5, 7, 35–37). These contrasts suggest that implementation strategies in the region should pair education with investments in interoperability, standards, and governance.

Participants emphasized the importance of establishing robust ethical and legal frameworks, ensuring data privacy, and achieving interoperability between AI systems and existing healthcare infrastructure. There was broad consensus on the need to integrate AI education into medical curricula to foster collaboration between clinicians and technology developers. While AI holds significant promise, especially in diagnostics and workflow optimization, participants stressed that it must complement, rather than replace, clinical judgment and the human aspect of care. Equitable access, affordability, and inclusive stakeholder engagement were identified as foundational to the responsible and sustainable implementation of AI in hepatology. These findings align with reports exploring regional perspectives on AI deployment in healthcare systems. Persistent technical issues such as algorithmic bias, overfitting, and limited generalizability are increasingly being addressed through region-specific datasets, stakeholder involvement, and rigorous validation protocols (4, 51). Evidence suggests that AI models trained on locally relevant data perform better and are more applicable to real-world practice (4, 51). In low- and middle-income countries (LMICs), leveraging locally sourced data, mobile health (mHealth) tools, and cloud-based infrastructures can help overcome limitations related to data quality and system capacity (49). Explainable AI is also gaining attention, enabling clinicians to better understand algorithmic outputs and enhancing trust (4). Ethical and legal challenges are being met through anonymization techniques, regulatory reforms, and international policy harmonization (49, 50). A notable example of global collaboration’s impact is a Google-led multicenter initiative that reduced false-positive and false-negative rates by 5–7%, thereby improving diagnostic accuracy and illustrating the value of targeted training and cross-sector partnerships (4). Region-aware governance and proactive stakeholder alignment further improve feasibility and uptake (52). Beyond diagnosis and staging, machine-learning models increasingly aim to forecast decompensation, mortality, and transplant-free survival, and to complement or recalibrate conventional scores (e.g., MELD) by leveraging non-linear interactions across clinical, imaging, and biomarker data. Early studies in ACLF and transplant pathways suggest potential for earlier risk identification, dynamic prioritization, and improved post-transplant outcome prediction; however, prospective validation and equity checks are essential before routine use (53, 54).

This study presents several notable strengths. It is the first comprehensive assessment of AI integration into hepatology practice and research across the MENA region, offering valuable regional insight where limited data currently exist. The study achieved a high response rate of 82.8%, with representation from 17 countries, enhancing its generalizability across diverse clinical settings. The use of a mixed-methods approach, combining quantitative data with thematic analysis of open-ended responses, allowed for an in-depth understanding of hepatologists’ perceptions, challenges, and recommendations. The survey instrument underwent pilot testing to ensure clarity and content validity, and the discussion is well-grounded in global literature, providing a contextual interpretation of the findings. These findings offer a foundation for future longitudinal and interventional research, emphasizing the urgent need for structured training, regulatory guidance, and strategic investment to facilitate the responsible integration of AI into hepatology practice.

This study has some limitations. First, reliance on self-reported perceptions may introduce recall, perception, and social-desirability bias; although anonymity and voluntary participation were used to mitigate this, reported use and trust may be over- or underestimated. Second, the cross-sectional design precludes causal inference and assessment of temporal trends; longitudinal follow-up is warranted to track changes in adoption and readiness. Third, the non-probability convenience/snowball recruitment may have introduced selection bias toward clinicians with stronger views about AI, limiting generalizability beyond our sample. Fourth, country-level numbers were small for several settings; accordingly, between-country comparisons are descriptive only and should be interpreted cautiously. Fifth, institutional readiness and AI use were not independently verified against records, and we did not collect objective clinical outcomes or detailed performance characteristics of specific AI tools. Given the sampling design and sparse strata, we prespecified a descriptive analytic approach and did not perform inferential hypothesis testing to avoid overstating population-level inference. Finally, missing data were handled by case-wise omission and some items permitted multiple responses; thus, denominators vary and percentages may not sum to 100%. Despite these constraints, the study’s multinational scope, high response rate, and mixed-methods design provide foundational, region-specific insights to guide policy, training, and implementation strategies for AI in hepatology.

AI adoption in MENA hepatology is characterized by high interest but limited routine use. Priority actions include (i) region-appropriate governance with clear accountability and human-in-the-loop safeguards; (ii) data standards and interoperability to enable integration with EHR/EMR and imaging systems; (iii) scalable training pathways—curricular integration, workshops, and simulation; and (iv) equitable access supported by local validation on MENA datasets. Coordinated efforts by ministries, professional societies, and academic centers are essential to translate optimism into safe, routine clinical benefit.