Introduction
Artificial intelligence (AI) has become a transformative force in modern healthcare, reshaping clinical practice, biomedical research, and healthcare administration. Through machine learning (ML), deep learning (DL), and natural language processing (NLP), AI systems can improve diagnostic accuracy, support predictive analytics [
1–
3], personalise treatment planning, and increase efficiency within healthcare systems. Notable applications include diagnostic imaging, predictive disease modelling, precision medicine, robotic-assisted surgery, and clinical decision support. Recent research indicates that AI-based models often outperform conventional algorithms in radiology and diagnostic imaging, enabling more accurate detection of lung, breast, and cardiovascular diseases. ML systems are widely used for disease-risk prediction and outcome prediction in chronic conditions, including diabetes and cardiovascular disease, whereas DL methods support automated feature extraction from medical images and genomic data in precision healthcare. At the same time, NLP approaches enable analysis of unstructured clinical data contained in electronic health records (EHRs) and the medical literature, thereby improving patient stratification and knowledge discovery. Applications of AI in surgery, particularly through surgical robotics, further demonstrate its clinical relevance by improving procedural precision and shortening recovery time. Alongside these developments, ethical and regulatory frameworks have received increasing attention in response to concerns regarding transparency, fairness, accountability, and patient safety in clinical AI [
4,
5]. This progress has been driven by the rapid expansion of biomedical data, increasing computational power, and the global shift toward digital health systems, all of which have accelerated AI adoption in healthcare over the past decade. Despite these advances, important limitations remain, including data-integration challenges, algorithmic bias, limited explainability of complex models, and inequitable access to AI technologies worldwide [
6]. Although previous reviews have examined individual AI methods or single clinical applications, relatively little research has integrated large-scale bibliometric mapping with comparative narrative synthesis to critically evaluate research patterns, collaboration networks, and methodological developments across AI methods in healthcare. In this paper, the authors present a comprehensive bibliometric and narrative review of 986 peer-reviewed articles published between 2015 and 2025 to address this gap. By combining quantitative bibliometric indicators with qualitative synthesis, this review identifies publication trends, global and institutional collaboration networks, prevalent and emerging AI methods, and major application areas in healthcare. It also examines AI-driven innovation in relation to ethical principles, sustainability, and global health equity, while outlining barriers and opportunities for the future development of responsible and scalable AI integration [
7]. These findings may inform researchers, policymakers, and healthcare stakeholders seeking to promote the effective, safe, and equitable implementation of AI in healthcare. Despite the rapid growth of AI-based healthcare research, existing reviews have focused primarily on individual methods, single applications, or qualitative overviews. Few studies have combined large-scale bibliometric mapping with comparative analysis of ML, DL, and NLP while also examining global collaboration networks and sustainability-related alignment. This review seeks to address that gap by integrating quantitative bibliometric analysis of 986 publications with guided narrative synthesis and by providing a reproducible, comparative, and policy-relevant overview of AI techniques in healthcare.
Contribution & Novelty
(1) Combines bibliometric mapping and narrative synthesis to provide a comprehensive evaluation. (2) Uses a comparative lens to assess ML, DL, and NLP across diagnostic imaging, predictive medicine, precision healthcare, and robotics. (3) Emphasizes collaboration by mapping global co-authorship and institutional networks and highlighting synergies among academia, industry, and healthcare. (4) Examines how AI in healthcare aligns with the United Nations Sustainable Development Goals (SDGs), with particular attention to fairness, ethics, and global health impact. (5) Proposes a future roadmap by identifying emerging areas for responsible adoption, including federated learning, explainable AI, and bias reduction (
Figure 1).
Objectives and Scope of the Study
This study aims to provide a comprehensive and comparative assessment of AI applications in healthcare between 2015 and 2025 using an integrated bibliometric and narrative synthesis approach. Specifically, it seeks to: (1) analyze publication growth, geographic distribution, and collaborative research networks; (2) compare major AI techniques—ML, DL, and NLP—across key healthcare applications, including diagnostic imaging, predictive medicine, precision healthcare, and robotic surgery; (3) identify emerging themes, ethical challenges, and sustainability-aligned innovations; and (4) propose data-driven future research directions to support responsible, equitable, and scalable AI adoption in healthcare systems.
Significance
AI techniques have the potential to transform healthcare by improving diagnostic performance, personalizing treatment plans, and increasing operational efficiency. Numerous studies have examined the impact of AI technologies in healthcare [
8]. AI-driven tools may improve prognosis, diagnosis, and treatment planning. In the future, AI is likely to become an increasingly important component of healthcare services and to influence a wide range of treatments. AI refers to computer-based systems, including software and hardware, that can perform certain tasks, solve problems, and make decisions without direct human involvement [
9,
10]. AI encompasses a wide range of approaches and applications, including genetic algorithms, neural networks, ML, and pattern recognition [
11]. The significance of this research lies in the following:
Improved Patient Outcomes
AI can be applied in healthcare to predict risks and provide recommendations. The use of big data and AI may improve patient health, diagnostic performance, and predictive capacity [
12]. Because AI can rapidly and accurately process and analyse complex medical data, it may support earlier diagnosis, stronger predictive analytics, and more effective therapeutic interventions, thereby improving patient care and outcomes.
Advances in Innovation
AI-related technologies may transform patient care and outcomes through their application in multiple areas of cancer care, including drug research and development, early diagnosis and screening, and drug discovery [
13]. This study examines how AI can advance medical imaging, drug discovery, and robotic-assisted practice, thereby promoting medical innovation.
Collaborative Synergy
AI has introduced new opportunities in healthcare and has transformed the delivery of medical activities and patient care. This paper examines the interactive relationship between AI and healthcare, with particular attention to the disruptive potential of AI technologies in the medical sector [
14,
15]. By mapping multidisciplinary relationships among academia, industry, and healthcare providers, the study highlights collaboration and resource sharing as important mechanisms for addressing global healthcare challenges and workforce shortages.
Ethical and Technical Challenges
This research contributes to the literature by synthesizing evidence on AI applications in healthcare. It also provides a detailed description of the ethical and governance issues that stakeholders seeking to implement AI in healthcare should consider. According to the review, ethical concerns, algorithmic bias, and data-integration challenges must be addressed to ensure that AI applications in healthcare are fair, transparent, and sustainable.
Future Directions
The paper provides stakeholders with a roadmap for realizing the promise of AI in healthcare by identifying emerging trends and opportunities and by addressing related sociotechnical barriers.
Methodology
Materials and Methods
This research used an integrated bibliometric and narrative review approach to examine studies on AI in healthcare published between 2015 and 2025. PubMed was selected as the primary data source because it provides broad coverage of peer-reviewed biomedical and healthcare literature.
Table 1 presents the detailed search strings used in the structured search, which combined keywords such as artificial intelligence, machine learning, deep learning, natural language processing, healthcare, diagnostic imaging, predictive analytics, precision medicine, and robotic surgery. Peer-reviewed English-language articles involving empirical, modeling, or systematic analysis of AI applications in healthcare were included. Excluded articles comprised of non-healthcare AI studies, editorials, commentaries, conference abstracts, and non-English publications. Article selection adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines (
Table 2). It included duplicate removal, title and abstract screening, and full-text evaluation, yielding a final dataset of 986 articles. R ver. 4.3.2 (R Foundation for Statistical Computing), RStudio (Posit PBC), and VOSviewer (Centre for Science and Technology Studies) were used to analyze publication patterns, citation trends, keyword co-occurrence, collaboration networks, and thematic clustering; the analytical tools and their functions are summarized in
Table 3.
Data Summary
A total of 986 articles on AI in healthcare published between 2015 and 2025 were reviewed across 399 journals and other academic sources. The dataset included 9,655 contributing authors, with an average of 12 co-authors per document and an international collaboration rate of 37.83%, indicating a highly multidisciplinary and globally connected research community (
Table 4). The annual publication growth rate was 8.1%, and the median document age was 4.47 years. Bibliometric analysis identified 4,453 keywords representing diverse research topics, with ML, DL, NLP appearing frequently in studies related to diagnostic imaging, predictive analytics, precision medicine, and robotic-assisted surgery. Publication distribution, co-authorship networks, and topic clusters were identified through visualization and network mapping using R ver. 4.3.2, RStudio, and VOSviewer, whereas narrative synthesis clarified patterns in the effects of AI on patient care, biomedical innovation, and operational efficiency. Challenges related to algorithmic bias, data integration, transparency, and patient privacy were also examined. The findings were further contextualized through cross-referencing with World Health Organization (WHO) and Food and Drug Administration (FDA) reports and with expert peer-reviewed literature, suggesting that AI in healthcare is a growing, collaborative, and multidisciplinary research field with both substantial opportunities and important ethical and practical challenges. The study selection process followed the PRISMA 2020 guidelines. The initial PubMed search identified 1,243 records published between 2015 and 2025. After removal of 143 duplicate records, 1,100 articles remained for title and abstract screening. During screening, 114 records were excluded for not meeting the study scope. Subsequently, 986 full-text articles were assessed for eligibility. All studies satisfied the inclusion criteria, resulting in a final dataset of 986 publications included in the bibliometric and narrative analyses.
Transparency in Analysis
This paper aims to maintain transparency by clearly presenting the methodology used to examine AI in healthcare, including bibliometric analysis, case-based evidence, and comparative analysis. The data sources are identified and include bibliometric records, published literature, and applied case examples. Collaborative networks were mapped using VOSviewer and Gephi (Gephi Consortium), and key AI techniques, including ML, DL, and NLP, were examined comparatively. Major areas of analysis, such as collaboration, AI applications, and ethical concerns, are clearly delineated. These findings were supported through cross-referencing with credible sources, including WHO and FDA materials, and through consideration of peer-reviewed literature. Ethical issues are explicitly addressed, and future opportunities for equitable AI collaboration are discussed. Overall, the study presents a clear, comprehensive, and structured analysis. The article-selection and screening procedure followed PRISMA principles to support transparency and reproducibility. Initially, 1,243 records were identified through a comprehensive PubMed search covering the years 2015 to 2025 using keywords including artificial intelligence, machine learning, deep learning, natural language processing, healthcare, precision medicine, predictive analytics, and robotic surgery. After duplicate removal and initial title and abstract screening, 257 articles were excluded because of duplication or lack of relevance. The remaining records underwent full-text review, and 986 papers were determined to meet the inclusion criteria. Sixty-eight articles were excluded because they were editorials, were unrelated to healthcare AI, or lacked sufficient methodological clarity.
Results
Bibliometric Parameters and Analytical Settings
To extract and analyze publication trends, collaboration networks, and topic structures, the bibliometric analysis was conducted using R ver. 4.3.2, RStudio, and VOSviewer. PubMed served as the primary data source, and the study parameters included authorship, institutional affiliations, year of publication, country of origin, citation counts, and author keywords. To improve analytical rigor, the minimum keyword-occurrence threshold was set at 5, and the minimum co-authorship link strength was set at 3. Documents with fewer than 10 citations were excluded from citation mapping. Theme clusters were identified using the VOSviewer modularity-based clustering algorithm (Louvain method), which applied association-strength normalization to balance links between frequently and infrequently occurring terms. Keywords were standardized through synonym merging (e.g., AI and artificial intelligence), removal of stop words, and lemmatization to improve comparability across studies. The resulting visualizations included keyword co-occurrence networks, author collaboration networks, and a topic-progression heatmap, providing an overall view of the research landscape from 2015 onward.
Bibliometric Findings
AI has advanced personalized healthcare applications, and research activity in this field has expanded alongside the growing volume and diversity of health-related data [
22]. This study used visual analytics, and bibliometric visualizations were generated in RStudio ver. 4.3.2 with VOSviewer. The visual elements used to improve the clarity and depth of the findings included the following.
Table 5 summarizes the descriptive characteristics of the final dataset comprising 986 included studies, consistent with the PRISMA selection process.
Co-authorship networks
To demonstrate interdisciplinary and cross-national research connections, particular attention was given to the international collaboration rate of 37.83%.
Citation maps
Citation mapping helped identify influential authors and their networks, including clusters centered on LeCun (DL), Esteva (medical imaging), and Topol (AI ethics).
Thematic cluster visualization
Figure 2 presents 6 major domains—diagnostic imaging, predictive medicine, precision healthcare, clinical text mining, surgical innovation, and collaborative networks—thereby adding interpretive depth to the findings. The inclusion of such graphical representations is important because it makes the bibliometric and comparative analyses not only statistically supported but also visually interpretable.
Geographical Distribution
The United States, China, and the European Union accounted for the highest levels of research output, supported by strong funding programs and policy initiatives. India, Republic of Korea, and the Middle East also emerged as growing contributors with increasing global engagement.
Institutional and Author Networks
Co-authorship and institutional analyses showed that major universities, hospitals, and research institutes were actively engaged in collaborative activity with commercial partners. Prominent funding bodies included the National Institutes of Health (US), the National Natural Science Foundation of China, and Horizon Europe.
Keyword Co-occurrence and Clustering
Bibliometric analysis identified 6 clusters of research activity (
Table 1). These clusters reflect major thematic orientations in healthcare AI and illustrate developments at the frontiers of research.
Citation Metrics
Highly cited publications included studies on convolutional neural networks (CNNs) for imaging, NLP for EHR mining, and surgical robotics. Many of the most highly cited papers were published in journals focused on medical informatics, biomedical engineering, and applied AI across disciplines.
Consolidation and interpretation of topic clusters
VOSviewer was used to perform thematic clustering based on keyword co-occurrence analysis, generating 6 aggregated clusters representing distinct but related areas of AI application in healthcare. These clusters were identified on the basis of semantic similarity and keyword co-occurrence frequency and were considered to represent the thematic structure of the field appropriately. The 6 main clusters were as follows: (1) diagnostic imaging, focused on the application of DL and CNNs to radiological interpretation; (2) predictive medicine, focused on big-data-driven models for chronic disease-risk prediction; (3) precision healthcare, (4) clinical text mining, dominated largely by NLP models such as Bidirectional Encoder Representations from Transformers (BERT) for processing unstructured EHRs; and (5) surgical innovation, focused on automation and robotics (
Table 1). These clusters were then interpreted through a combination of bibliometric strength indicators, including link weight and centrality, and narrative synthesis, showing how AI methods interact with both the clinical and operational dimensions of contemporary healthcare.
Bibliometric indicators supporting the analysis
The impact, structure, and collaborative dynamics of AI research in healthcare were evaluated using a range of bibliometric indicators. The total corpus had an h-index of 67, indicating substantial and sustained citation impact. Intellectual visibility was also high, with an average citation rate of 12.3 citations per item. Co-citation analysis identified foundational authors such as LeCun (DL), Esteva (medical imaging), and Topol (AI ethics) as central to the field’s intellectual structure. Cluster-centrality analysis identified diagnostic imaging and predictive medicine as particularly influential research areas, as both showed high betweenness centrality and served as bridges between technical and clinical domains. The collaboration index, calculated as the average number of co-authors per article, was 12.0, reflecting strong multidisciplinary participation. In addition, 37.83% of all included articles involved international co-authorship, underscoring the global and collaborative nature of healthcare AI research. Together, these metrics support the maturity, connectedness, and resilience of the field over the study period.
Trends in AI applications in healthcare
According to the Institute of Medicine (IOM) of the National Academies of Sciences, preventable healthcare errors result in 44,000 to 98,000 deaths annually in the United States. The IOM has criticized the healthcare sector for its inability to adopt new technologies in a consistently safe and effective manner [
23,
24]. In recent years, AI has shown considerable potential for improving the quality and efficiency of diagnosis, predicting patient outcomes, and personalizing treatment plans [
25]. AI applications in healthcare have advanced substantially, particularly in diagnosis, treatment planning, and healthcare management. ML algorithms may support faster and more accurate diagnosis, whereas AI systems may also help tailor treatment strategies to individual patients and improve clinical outcomes. Healthcare management has similarly benefited from AI through improvements in resource allocation and process optimization. Interest in this field has been reflected in the rapid increase in publications and patents related to AI in healthcare. Emerging applications are also beginning to influence areas such as mental health diagnostics, rural healthcare accessibility, and early-stage disease prediction, suggesting broadening use across healthcare domains.
Enhanced comparative analysis through ML, DL, and NLP
A comparative analysis of ML, DL, and NLP highlights their distinct yet complementary roles in healthcare. ML is widely used in predictive modeling and patient-risk assessment because of its interpretability for structured data, although it is less effective for highly complex or unstructured information. DL dominates diagnostic imaging, robotics, and radiomics because of its high accuracy and automated feature extraction, despite limitations related to data requirements and limited transparency. NLP is essential for analyzing unstructured clinical material, such as EHRs and the medical literature, because it enables extraction of meaningful insights despite challenges related to linguistic variation and data quality. Collectively, these approaches suggest that integrated or hybrid AI models combining DL’s analytical power, ML’s interpretability, and NLP’s language-processing capabilities may provide a more comprehensive framework for healthcare innovation (
Table 6).
Thematic clusters
AI is now widely used to support diagnosis and administration in healthcare. Since the coronavirus disease 2019 (COVID-19) pandemic, interest has increased in the deployment of AI and robotics to protect healthcare personnel from exposure during patient care [
26]. The clustering analysis identified 6 major research themes: (1) Diagnostic imaging: AI models, including CNNs and other DL methods, are used to analyze magnetic resonance imaging and computed tomography scans to improve disease detection and radiological workflows. (2) Predictive medicine: Predictive analytics and big-data approaches are improving risk prediction, particularly for chronic diseases such as diabetes. (3) Precision healthcare: Integration of genetic and clinical data supports tailored therapy and precision medicine. (4) Clinical text mining: NLP approaches, including BERT-based models, can analyze unstructured EHRs and medical literature to improve patient classification and knowledge discovery. (5) Surgical innovation: Robotics and automation improve surgical precision and recovery. (6) Collaborative networks: Research on digital healthcare ecosystems highlights resource sharing, distributed decision-making, and innovation in older-adult care.
Trends, comparative analysis, and biomedical innovation
The evolution of AI research in healthcare shows a progression from early emphasis on diagnostic imaging and predictive analytics (2015–2018), to greater use of NLP and robotics during the COVID-19 period (2019–2021), and more recently (2022–2025) to areas such as mental health, rural care, and rare-disease diagnosis (
Figure 2). DL has outperformed traditional ML in many imaging tasks, NLP remains essential for analysis of unstructured clinical data, and hybrid models are increasingly being developed for more comprehensive analysis. These technologies are driving biomedical innovation by accelerating drug development, supporting genomics-based personalized medicine, and improving surgical precision through adaptive robotics. AI may also improve epidemic prediction and resource allocation in healthcare, whereas partnerships among academia, industry, and healthcare organizations remain important for translating innovation into practice. The bibliometric findings also contribute substantially to comparative analysis of AI applications involving ML, DL, and NLP. The dataset comprised 986 peer-reviewed articles published between 2015 and 2025 across 399 academic sources, with an annual growth rate of 8.1% and a mean of 12 co-authors per paper, indicating strong global research collaboration. Of the 6 identified thematic clusters, diagnostic imaging and predictive medicine showed the highest betweenness centrality, consistent with the dominant roles of DL in imaging and ML in predictive analytics. The field’s intellectual maturity is further reflected in an h-index of 67 (
Table 7) and a mean citation rate of 12.3 citations per document. In addition, publications related to NLP increased markedly in 2020, likely reflecting rising interest in EHR mining and data analytics during the pandemic. These numerical findings support the complementary roles of ML, DL, and NLP in healthcare innovation. Recent articles also provide practical examples of advanced AI models such as federated learning and explainable AI. Federated learning has emerged as a privacy-preserving approach for training models across multiple healthcare institutions while maintaining data protection and improving the robustness of diagnostic systems. For example, studies [
27,
28] reported that federated learning improved breast cancer prediction and supported multicenter collaboration without direct data exchange. In contrast, explainable AI has been proposed to enhance clinical trust and decision transparency, particularly in radiology and precision medicine, by making algorithmic reasoning more interpretable to clinicians and patients. These examples illustrate the practical relevance and ethical promise of AI in real-world healthcare settings. Beyond summarizing trends, the review also emphasizes methodological rigor, generalizability, and therapeutic relevance through critical analysis of high-impact studies. The included research was evaluated in relation to dataset diversity, external validation, reporting transparency, and real-world testing outcomes. Although some studies were methodologically innovative, many still require further validation and clinical testing. The review therefore categorizes the literature into 3 stages of maturity—basic methodological studies, translational pilots, and scalable deployments—with common strengths including rigorous validation and transparency, and common limitations including sample bias and limited clinical-trial evidence. This critical perspective underscores the importance of fairness audits, future-oriented analysis, and implementation-focused research in moving AI in healthcare beyond experimental success toward evidence-based clinical practice.
Future directions and opportunities
Future developments in healthcare AI should emphasize emerging applications such as mental health diagnosis and rare-disease detection. AI may also substantially improve personalized medicine through integration of imaging, genetic, and clinical data to identify individualized risks, optimize treatments, and support patient monitoring. Infectious diseases such as HIV have also become targets for AI-enabled prediction of treatment outcomes, detection of adherence risks, and identification of resistance mutations to guide therapy selection. These applications may enable more precise treatment, earlier intervention, and better-targeted public health strategies. However, additional efforts in clinical validation, data diversity, and equity remain necessary to ensure that these benefits are distributed fairly across populations (
Table 3). AI-based personalized treatment planning may support advances in these areas. Research in these domains may improve early detection, patient outcomes, and the capacity of healthcare systems to address existing gaps. Emerging approaches such as federated learning also enable privacy-preserving AI training by allowing institutions to collaborate without disclosing sensitive patient data, thereby improving model performance while protecting confidentiality. In addition, explainable AI may play an essential role in strengthening confidence in AI-driven healthcare decisions by making model outputs more visible and interpretable to both clinicians and patients. To promote equitable AI-based healthcare solutions globally, future strategies must include broader access to AI technologies in under-resourced settings, mitigation of bias through more diverse data representation, and the development of global partnerships for ethical AI systems. Governments, technology companies, and healthcare institutions should work together to reduce disparities and promote health equity worldwide.
The implications of AI breakthroughs in healthcare are substantial because they may help address fundamental challenges, improve patient outcomes, and stimulate progress in medical practice. AI algorithms emulate aspects of human learning, synthesis, analysis, generalization, and problem-solving and include approaches such as NLP, ML, DL, and large language models [
29]. The performance of an AI system depends heavily on both data volume and data quality. To evaluate the full potential of AI algorithms, large and diverse datasets drawn from multiple healthcare systems are needed [
30]. The significance of these findings can be summarized as follows: (1) Improved patient care: AI may support earlier disease detection, more personalized treatment, and better outcomes for patients with rare diseases. (2) Bridging healthcare gaps: AI may improve healthcare access in underserved areas through mobile health technologies and federated learning while maintaining data confidentiality. (3) Trust and ethical AI: Explainable AI may improve transparency and reduce bias, thereby supporting more equitable decision-making across populations. (4) Public health and crisis preparedness: AI may strengthen health-system resilience, improve preparedness, and support epidemic detection through more effective use of healthcare resources. (5) Global AI standardization: International collaboration may support ethical AI, regulatory alignment, and more equitable healthcare solutions worldwide.
How AI in Healthcare Aligns with Sustainability Goals
Bibliometric evidence suggests that healthcare AI research has increasingly converged on sustainability-oriented themes, particularly after 2020, as reflected by growing keyword co-occurrence with terms such as ethics, equity, privacy, and global health access. Applications of AI in healthcare, including medical imaging analysis, predictive modeling, and personalized medicine, Patient-Reported Outcomes & Value Evidence, may improve healthcare delivery, expand access to high-quality care (
Figure 3), and reduce inequities [
31,
32]. Direct AI applications in healthcare may also support several United Nations SDGs, including improved access to quality healthcare, equitable treatment, and more efficient resource use. Emerging bibliometric clusters also indicate growing interest in federated learning, reflecting an emphasis on privacy-preserving model training. Federated learning may help address major concerns related to patient privacy and data security because it avoids direct data exchange across institutions while allowing decentralized learning, thereby supporting ethical AI use and aligning with SDG 3, which emphasizes safe, inclusive, and high-quality healthcare. The increasing frequency of keywords related to NLP likewise suggests growing interest in text-based clinical intelligence and patient communication. However, the relatively limited use of NLP in mental health research points to a valuable avenue for future work, particularly the development of more accessible and scalable tools for early psychological screening, patient engagement, and support in digitally underserved populations. The following points connect these findings to global sustainability goals.
AI for universal health coverage
AI may strengthen universal health coverage by supporting early diagnosis, personalized treatment, and improved patient outcomes. It may also help address gaps in mental health and rare-disease care while enhancing population-health surveillance for epidemic prevention and health-system resilience.
Federated learning and ethical AI
Federated learning and ethical AI frameworks may support secure AI research by protecting patient information and promoting compliance with legal and regulatory requirements [
33]. Explainable AI may also improve reliability and transparency in medical decision-making, thereby increasing trust in AI-based care. In addition, AI-informed policies may strengthen healthcare governance and security while promoting ethical and responsible AI use.
AI for equitable
AI may improve equitable healthcare access through telemedicine and mobile applications that extend care to underserved communities. Reducing bias in AI systems may also promote fairer treatment across diverse populations and help reduce disparities in healthcare delivery.
AI for resource efficiency and climate
Predictive AI may improve sustainability by strengthening allocation and supply-chain management of medical resources. AI-based disease surveillance may also improve preparedness for health emergencies. In addition, remote AI-enabled healthcare may reduce emissions and environmental burden by decreasing unnecessary travel and resource waste [
34].
Discussion
The adoption of AI in healthcare may support sustainability by improving patient care, increasing resource efficiency, and reducing inequities in access. Through predictive analytics, personalized medicine, and AI-based diagnostics, earlier disease detection and more targeted treatment planning may improve patient outcomes while supporting more efficient use of medical resources. Federated learning and ethical AI approaches may also strengthen data privacy, transparency, and trust in decision-making. Mobile health solutions and telemedicine may help reduce healthcare disparities while also decreasing the environmental burden associated with unnecessary travel and inefficient resource use. Together, these developments suggest that AI may contribute to a more sustainable, efficient, and inclusive global healthcare system. The potential of AI to address major healthcare challenges and support sustainable development across multiple domains is substantial. AI may improve healthcare delivery through earlier diagnosis, personalized treatment strategies, and predictive analytics for population-health programs [
29]. Although AI research and healthcare research have traditionally differed in structure and emphasis, the 2 fields are becoming increasingly interconnected. AI research focuses primarily on scalable algorithms, computational models, and automated analytical methods such as ML, DL, and NLP. In contrast, healthcare research remains patient-centered and grounded in clinical trials, epidemiology, and evidence-based treatment, with strict ethical and regulatory requirements. The use of AI in healthcare bridges these domains by enabling data-driven innovation without abandoning the human-centered principles of care. Manual administrative practices in hospitals may also hinder efficient decision-making in healthcare systems. AI technologies have therefore been introduced into healthcare not only to address existing operational challenges, but also to create new opportunities through improved data processing and analysis [
32,
33]. From a sustainability perspective, AI may support universal health coverage by enabling earlier diagnosis, personalized medicine, and remote healthcare access in underserved settings. Ethical frameworks, federated learning, and explainable AI may help ensure that implementation is secure, transparent, and equitable, thereby strengthening stakeholder confidence. AI-supported optimization of medical supply chains and predictive analytics may also improve system efficiency and climate resilience. Taken together, these findings underscore the potentially transformative role of AI in promoting sustainable and equitable healthcare worldwide.
Ethical and Practical Challenges in AI-Driven Healthcare
The use of AI in healthcare raises major ethical and practical challenges, including algorithmic bias, data privacy concerns, limited transparency, legal ambiguity, and infrastructural constraints. Bias often arises from unrepresentative datasets and may lead to unequal outcomes across populations. In addition, fragmented EHR systems and nonstandardized data formats make model training and validation more difficult. Clinician trust depends heavily on both competence and accountability, making interpretable and clinician-in-the-loop systems especially important. Practical implementation is further hindered by regulatory uncertainty and limited computing resources. To address these challenges, global standards such as Institute of Electrical and Electronics Engineers (IEEE) and WHO guidance recommend privacy-preserving approaches, including federated learning, fairness audits, transparent reporting, and structured governance. Emerging best practices emphasize diverse data inclusion, clinician engagement, interoperable data standards, and scalable infrastructure to support AI in healthcare that is more ethical, transparent, and sustainable. Several reviewed studies also highlighted concrete ethical concerns. For example, DL models for imaging that were trained primarily on Western datasets showed lower diagnostic accuracy in underrepresented populations, raising concerns about algorithmic bias. NLP-based EHR-mining studies also reported privacy risks associated with potential re-identification, especially when multi-institutional datasets were used. In addition, black-box DL systems in radiology were reported to undermine clinician trust because of limited explainability, prompting calls for more robust explainable-AI frameworks in clinical decision support.
Future Research Directions
According to the bibliometric patterns observed in this review, the future of AI in healthcare may shift away from an exclusive focus on imaging-based applications and place greater emphasis on less represented areas such as mental health, where AI-based screening, monitoring, and intervention tools remain underused despite increasing demand. Greater geographic diversification is also needed, because current research output remains concentrated in high-income settings; stronger participation from low- and middle-income countries is necessary, particularly through resource-efficient and context-sensitive AI solutions. In addition, although many highly cited studies are methodologically innovative, an important gap remains in large-scale clinical validation, highlighting the need for real-world implementation studies and longitudinal analysis. Privacy-preserving strategies such as federated learning should be expanded to support cross-institutional collaboration without compromising patient confidentiality, thereby enabling more ethical and scalable healthcare AI systems (
Table 8). At the same time, integration of explainable AI is essential, particularly in high-risk settings such as oncology and radiology, to improve transparency, trust, and clinical uptake. Taken together, these priorities emphasize the need to align technological progress with ethical governance, international equity, and long-term healthcare sustainability.
