Front. Public Health Frontiers in Public Health Front. Public Health 2296-2565 Frontiers Media S.A. 10.3389/fpubh.2025.1656603 Review AI-driven transformation of precision medicine: a comprehensive narrative review of key application areas, emerging paradigms, and future directions ZengQin 1 2 Methodology Conceptualization Writing – review & editing Writing – original draft Investigation HuangCheng 2 3 Formal analysis Validation Data curation Writing – review & editing ZhuJun 2 3 * † Project administration Supervision Writing – review & editing Department of Pediatric Gastroenterology Nursing, West China Second University Hospital, Sichuan University, Chengdu, Sichuan, China Key Laboratory of Birth Defects and Related Diseases of Women and Children (Sichuan University), Ministry of Education, Chengdu, Sichuan, China National Office for Maternal and Child Health Surveillance of China, National Center for Birth Defect Surveillance of China, Department of Pediatrics, West China Second University Hospital, Chengdu, Sichuan, China Correspondence: Jun Zhu, zhujun028@163.com

ORCID: Jun Zhu, orcid.org/0009-0003-3329-9694

 08 01 2026 2025 13 1656603 30 06 2025 27 11 2025 08 12 2025 Copyright © 2026 Zeng, Huang and Zhu. 2026 Zeng, Huang and Zhu https://creativecommons.org/licenses/by/4.0/ This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms. Objectives

This study aims to elucidate the pivotal role of Artificial Intelligence (AI) in driving the transformation of precision medicine, comprehensively analyzing how it reshapes healthcare systems from traditional diagnosis and treatment paradigms into personalized health management ecosystems.

 Methods

A comprehensive narrative review was conducted to systematically synthesize and critically evaluate the innovative applications, paradigm shifts, and future prospects of AI across the entire precision medicine value chain. A comprehensive literature search was performed across multiple databases up to April 30, 2025, with a focus on the clinical implementation and breakthroughs of technologies such as deep learning (DL), machine learning (ML), and natural language processing (NLP).

 Results

AI technologies have significantly enhanced the accuracy and efficiency of disease diagnosis through medical image analysis, genomics, and multimodal data fusion. At the treatment level, AI enables the development of personalized therapeutic plans and drug dosing optimization, while revolutionarily accelerating the drug development pipeline from discovery to clinical trials. Integrated with wearable devices and telemedicine platforms, AI facilitates full-cycle health monitoring. However, the clinical translation of AI faces challenges, including an uneven evidence base, insufficient model generalizability, and ethical concerns regarding data privacy, algorithmic fairness, and interpretability.

 Conclusion

AI is a key driver of paradigm shift in precision medicine. To address existing challenges, future efforts should focus on generating more robust clinical evidence, adopting technologies like federated learning to ensure data privacy, and promoting the human-centered, collaborative framework of Symbiotic AI (SAI). By establishing sound ethical and governance structures, the deployment of AI technologies can be ensured to be not only efficient and advanced but also equitable and trustworthy, ultimately paving the way for an intelligent and inclusive healthcare ecosystem.

 artificial intelligence deep learning healthcare ethics personalized treatment precision medicine remote monitoring symbiotic artificial intelligence The author(s) declared that financial support was not received for this work and/or its publication. section-at-acceptance Digital Public Health Introduction

In the wave of paradigm shifts in medicine, Precision Medicine is reshaping the modern medical system with its revolutionary concept of individualized diagnosis and treatment. This model integrates multidimensional data such as genomics, phenomics, and environmental exposure data (1), constructing diagnostic and therapeutic decision-making systems based on patient-specific biomarkers. The core essence lies in breaking through the limitations of traditional “population-based treatment” and achieving a paradigm shift from disease treatment to health management (2). Notably, groundbreaking advances in Artificial Intelligence (AI) are accelerating this transformation. Technologies such as imaging omics analysis driven by Deep Learning (DL) (3), clinical decision models powered by Machine Learning (ML) (4), and electronic medical record mining supported by Natural Language Processing (NLP) (5) are forming a technological matrix that drives Precision Medicine, with a hierarchy spanning from foundational ML to advanced concepts like DL and Large Language Models (LLMs) (Figure 1).

The technical hierarchy of modern artificial intelligence. This hierarchy categorizes AI based on the model’s output (discrimination vs. generation). CNNs, Convolutional Neural Networks; GANS, Generative Adversarial Networks; VAES, Variational Autoencoders; SVM, Support Vector Machine; LogReg, Logistic Regression.

Flowchart illustrating the hierarchy of Artificial Intelligence, Machine Learning, and Deep Learning, leading to Discriminative AI and Generative AI. Discriminative AI includes image classifiers, prediction models, and traditional machine learning. Generative AI covers large language models, GANs, VAEs, and diffusion models.

The practice of Precision Medicine essentially forms a complex data feedback loop: from genomic sequencing data and multimodal imaging data to real-time physiological parameters monitored by wearable devices, the integration and analysis of massive heterogeneous data present a significant challenge to traditional healthcare systems (6). In this context, AI technologies demonstrate unique empowering value: at the diagnostic level, convolutional neural network-based imaging recognition systems can achieve sub-millimeter lesion localization (7); in treatment decision-making, reinforcement learning (RL) algorithms can generate personalized treatment plan recommendations (8). Zhang et al. (9) systematically illustrated how AI is transforming drug development, predicting it could reduce the development cycle by up to 70%. Furthermore, AI-driven smart monitoring systems have enabled real-time dynamic monitoring of patients’ vital signs (10–12). Currently, the penetration of AI is exhibiting significant depth: the introduction of Transformer architectures has made cross-modal healthcare data fusion possible (13); the maturity of Federated Learning (FL) frameworks effectively addresses the dilemma of healthcare data silos (14); a technique that allows multiple institutions to collaborate on model training without sharing sensitive data, ensuring patient privacy (15), and intelligent consultation systems based on LLMs are reshaping doctor-patient interaction patterns (16).

However, it is important to recognize that technological evolution is always accompanied by ethical challenges—issues such as healthcare data privacy protection, lack of algorithm interpretability, and insufficient model generalization ability have become key bottlenecks restricting the deep application of AI (17–19). Excitingly, emerging technologies such as blockchain-enabled distributed learning architectures and knowledge graph-enhanced Explainable AI (XAI) offer innovative pathways to address these challenges (20–22). XAI aims to make AI decision-making transparent and interpretable for clinicians, using techniques like visual analytics and knowledge graphs to foster trust in AI-driven insights (23). Simultaneously, AI technologies demonstrate unique advantages in processing complex biological information (24). Given the interdisciplinary, heterogeneous, and rapidly evolving nature of this field, a narrative review methodology is adopted. Evidence spans multiple domains - including computational algorithms, genomics, and emerging technologies like FL - making it challenging to apply systematic reviews that require homogeneous evidence types (25). In contrast, a narrative review offers the flexibility needed to synthesize these diverse findings and analyze AI’s transformative role across the precision-medicine ecosystem.

This paper systematically reviews the innovative applications of AI technologies across the entire Precision Medicine value chain. To ensure transparency, a comprehensive literature search was conducted across multiple databases, including PubMed, MEDLINE, PsycINFO, Web of Science, and EMBASE, from their inception to April 30, 2025. The search strategy focused on the intersection of “artificial intelligence” and “precision medicine,” incorporating technical keywords (deep learning, machine learning, natural language processing) and clinical terms (clinical decision support, personalized treatment, genomic biomarkers). Consistent with the narrative review methodology, literature selection was based on thematic relevance, methodological contribution, and academic impact. The initial search yielded several thousand records, which were screened to retain a few hundred core articles for full-text analysis. Inclusion criteria prioritized peer-reviewed studies directly related to AI in precision medicine, while excluding non-English and non-peer-reviewed publications. Although formal quality appraisal tools (e.g., GRADE) were not used, methodological transparency, sample robustness, and publication in high-impact journals were key selection factors to ensure the credibility and rigor of the synthesized evidence (26). A simplified flow diagram (Figure 2) visually summarizes the search and selection process.

Simplified flow diagram of the literature search and selection process. This flow diagram represents an illustrative overview of the narrative review process rather than a PRISMA-based systematic workflow.

Flowchart detailing the selection process for qualitative synthesis. It starts with records identified from database searches (PubMed, MEDLINE, PsycINFO, Web of Science, EMBASE). Next, duplicates are removed. Then, title and abstract screening occurs, followed by a full-text review of relevant articles, and finally the selection for qualitative synthesis.

By critically evaluating the strengths and limitations of existing technologies and incorporating cutting-edge trends, this review synthesizes a comprehensive overview of the AI-enabled technological landscape for Precision Medicine. Its primary aim is to provide theoretical support and practical reference for building an intelligent healthcare ecosystem. Furthermore, the review extends beyond technological synthesis to critically examine the varying levels of clinical validation-including performance benchmarks like AUROC and the need for more comparative studies and meta-analyses-across different medical specialties. A pivotal discussion involves the challenge of ensuring model generalizability across diverse healthcare settings, which is essential for the equitable and widespread adoption of AI in precision medicine.

 Technological empowerment: core application scenarios of AI in precision medicine

AI is driving profound transformations in the fields of precision medicine and nursing. Its applications span disease diagnosis, treatment planning, drug development, and patient monitoring, providing robust support for personalized and precise medical services. To establish a technical foundation for the subsequent discussion, Table 1 provides a comparative overview of the most prominent AI techniques in precision medicine, summarizing their key applications, strengths, and limitations, serving as a quick reference guide for understanding the technologies discussed herein.

Comparison of AI techniques in precision medicine.

Technique/framework Key applications Strengths Limitations
Convolutional Neural Networks (CNNs) Medical image analysis (e.g., MRI, CT scans), lesion detection, tumor classification High accuracy in image recognition, automated feature extraction Requires large labeled datasets, prone to overfitting, lacks interpretability
Generative Adversarial Networks (GANs) Synthetic medical image generation, data augmentation Enhances dataset diversity, useful for rare diseases, privacy-preserving Quality of synthetic data may vary, ethical concerns, potential for bias
Random Forest Genomic data analysis, disease prediction, biomarker identification Handles high-dimensional data, interpretable, robust to overfitting Struggles with imbalanced datasets, less effective for unstructured data
Bidirectional Encoder Representations from Transformers (BERT) Clinical text analysis, EHR mining, patient communication analysis Contextual understanding of text, improves NLP tasks, supports decision-making Requires large text datasets, computationally intensive, lacks interpretability
Deep Q-Networks (DQN) Personalized treatment optimization, dynamic treatment adjustments Handles complex decision-making, learns from feedback, real-time adaptation Requires high-quality real-time data, complex reward functions, suboptimal decisions possible
Multimodal Large Language Models (MLLMs) Integration of text, images, and clinical data, personalized treatment recommendations Holistic data analysis, improves diagnostic accuracy, supports multimodal data Requires high-quality multimodal datasets, interpretability challenges
Explainable AI (XAI)† Model interpretation†† Improves transparency Potential performance trade-offs

†XAI is a framework or technique to enhance interpretability. ††The interpretability strengths refer to XAI-enhanced models; native model interpretability varies by complexity.

Building upon these technologies, AI demonstrates significant impact across key clinical scenarios. Before each area is explored in depth in the following subsections, Table 2 serves as a consolidated evidence map, summarizing the key findings from the literature regarding AI’s efficacy in diagnosis, treatment, drug development, and health monitoring. Furthermore, Figure 3 summarizes the diverse applications of AI across different stages of precision medicine (from data collection to diagnosis, treatment, drug development, and monitoring) from a macro-process perspective, illustrating how AI technologies are integrated to enable AI-driven decision-making through data integration, clinical decision support, and enhanced patient engagement.

Synthesis of literature on AI’s role in precision medicine.

Domain Key findings Impact
Diagnosis AI, particularly CNNs, improves diagnostic accuracy in medical imaging (e.g., MRI, CT scans) and lesion detection Reduces misdiagnosis and missed diagnosis, enhances early detection of diseases like cancer
Treatment AI optimizes personalized treatment plans by analyzing genomic data, clinical information, and lifestyle factors Improves treatment efficacy, reduces side effects, and supports dynamic treatment adjustments
Drug development AI accelerates drug discovery by analyzing compound databases and predicting drug activity and toxicity Shortens development cycles and reduces costs
Health monitoring AI, combined with wearable devices, enables real-time health monitoring and early disease prediction Enhances proactive healthcare, reduces hospital readmissions, and improves patient outcomes
Clinical decision support NLP and AI-driven systems extract insights from EHRs and clinical notes, supporting evidence-based decision-making Improves diagnostic accuracy, treatment recommendations, and patient safety
Ethical challenges AI faces challenges in data privacy, algorithmic transparency, and fairness, particularly in diverse populations Requires robust ethical frameworks and XAI to ensure trust and equitable healthcare delivery

This table summarizes key findings from the reviewed literature, highlighting the impact of AI in precision medicine across different domains such as diagnosis, treatment, and drug development. It provides a concise overview of the evidence supporting AI’s role in transforming healthcare.

AI applications in precision medicine. This figure illustrates the workflow from data collection (genomic, imaging, wearable devices) to AI-driven decision-making, highlighting the integration of AI technologies (e.g., DL, ML, NLP) across different stages of precision medicine.

Flowchart depicting the stages of AI-driven precision medicine: Data Collection (genomic data, medical imaging, wearable device data), Diagnosis (multimodal data fusion, comprehensive profiling), Treatment (personalized plans, dosage optimization), Drug Development (screening, trial optimization, repositioning), Monitoring (real-time health, disease prediction), and Decision-Making (AI-powered support).

To enhance critical evaluation, this section incorporates multi-dimensional analytical perspectives when presenting each application - comparing different AI methodologies (e.g., Convolutional Neural Networks (CNNs) vs. Transformers, centralized vs. federated architectures), analyzing real-world performance limitations and underperformance cases, and addressing dataset bias and domain shift challenges - to achieve a more balanced and practical understanding of AI applications.

Optimization of diagnostic decision-making

This section explores how AI enhances diagnostic accuracy across data modalities, while also examining the technical and practical challenges that influence its clinical adoption.

Medical image analysis

DL, particularly CNNs, has become a cornerstone of medical image analysis. These models learn hierarchical visual features from CT, MRI, and X-ray images to perform tasks such as disease detection, quantification, and prognosis with high accuracy (6, 27–31).

A prominent example of their impact is in mammography for breast cancer diagnosis. DL models have demonstrated performance comparable to that of professional radiologists in detecting malignancies (32–34). This capability significantly enhances diagnostic accuracy and helps reduce human errors, including misdiagnosis and missed diagnosis.

A comparative analysis reveals a trade-off between model complexity and practical applicability. While CNNs are highly effective, their performance typically depends on large, annotated datasets, and they face challenges of overfitting and interpretability in high-stakes clinical settings (35–37). More recently, Transformer-based models have shown superior ability in capturing long-range dependencies and facilitating multimodal data fusion (38, 39). However, their substantial computational cost and even greater data demands make them challenging to deploy in resource-constrained environments (40). In contrast, CNN models remain computationally more efficient and are often better suited for deployment in scenarios with limited data or computational resources, such as in primary care clinics or on portable imaging devices (23).

 Multimodal image fusion

Multimodal Large Language Models (MLLMs) align and fuse images, text, and other clinical signals via cross-modal attention to support joint reasoning for diagnosis and prognosis (13, 41); building trustworthy systems demands high-quality curated multimodal datasets and careful attention to interpretability and transparency (42–44).

AI technology can integrate data from different imaging modalities (such as MRI and CT), providing more comprehensive and three-dimensional disease information. This is particularly important in brain tumor and lung nodule diagnoses, helping clinicians accurately locate lesions, assess the nature of abnormalities, and develop more scientifically grounded treatment plans (38, 45). Recent advancements in digital pathology have further expanded the role of AI in diagnostics. For instance, AI-based systems have been successfully applied to automate the analysis of histopathological images, reducing diagnostic time and improving accuracy. A recent study demonstrated that an AI-powered platform could accurately detect, grade, and quantify prostate cancer from biopsy samples, significantly outperforming traditional diagnostic methods. This advanced system improves diagnostic accuracy and reduces interobserver variability, offering the potential to revolutionize histopathological evaluation and enhance clinical management and risk stratification for prostate cancer (46).

Nevertheless, AI models often demonstrate poor generalization when applied across hospitals or devices, a phenomenon widely documented in studies that reveal significant performance drops due to domain shift (47). This is exemplified in CT imaging, where a model developed for COVID-19 diagnosis experienced a substantial decrease in accuracy when tested on data from a different country and scanner manufacturer, directly illustrating the clinical risks of dataset bias and vendor-specific training (48).

 Genomics and biomarker identification

AI technology, particularly ML, has become instrumental in mining genomic data to extract disease-related biomarkers from large-scale datasets. By analyzing individual genetic variations, ML algorithms can predict susceptibility to specific diseases, thereby laying a crucial foundation for early diagnosis and prevention (49). For instance, Random Forest ensembles are effective at handling high-dimensional genomic features to discover disease-associated variants and build polygenic risk predictors (50, 51). However, their utility is often constrained by class imbalance and inherent interpretability challenges (50, 52).

These methods have shown significant promise in complex disorders. A notable example is the application of AI models to identify genetic markers for Alzheimer’s disease, facilitating early intervention strategies. Miller et al. (53) recently demonstrated the potential of such approaches in Alzheimer’s genetics, underscoring their role in discovering novel markers and predicting individual risk.

A critical limitation, however, lies in the generalization of these models across diverse populations. The heterogeneity and high dimensionality of genomic data mean that AI models often face significant performance degradation when applied across different ethnic groups. Such algorithmic bias, where models trained on non-representative data fail to generalize, can exacerbate health disparities, as demonstrated in studies of clinical decision-making algorithms (54). In genomics, this is starkly evident in polygenic risk scores, where models trained predominantly on European-ancestry data show markedly reduced predictive accuracy in non-European populations, ultimately impacting the fairness and accuracy of early disease identification (55).

 Comprehensive clinical data analysis

The application of AI in precise diagnostics extends beyond imaging and genomics to encompass the comprehensive analysis of clinical data, laboratory results, and lifestyle information, thereby supporting more accurate individualized diagnoses. Specifically, Transformer-based clinical NLP models, such as Bidirectional Encoder Representations from Transformers (BERT) variants, excel at encoding contextual semantics within Electronic Health Records (EHRs) to extract phenotypes, medications, and clinical outcomes for downstream modeling and cohort construction (56–58).

These technologies have yielded significant real-world impact. For instance, by analyzing EHR data, AI can identify high-risk patients and enable early intervention, effectively reducing the incidence of severe diseases (59, 60). In the field of telemedicine, AI-driven diagnostic tools integrated into healthcare platforms provide real-time diagnosis and consultation for patients in underserved areas, effectively bridging geographical disparities in healthcare access (61).

However, reveals both the considerable promise and the accompanying challenges of these applications. The core strength lies in the capability to process vast, multi-source clinical data to enhance diagnostic accuracy and decision-making efficiency. Nevertheless, the effectiveness of these advanced models is highly dependent on large-scale, domain-specific corpora and substantial computational resources (57). More critically, the equitable deployment of AI faces systemic challenges. Algorithmic biases can perpetuate and amplify existing health disparities, as evidenced by studies on population health management algorithms (54). Concurrently, persistent digital literacy and technology access gaps between different regions and populations risk excluding the most vulnerable from the benefits of innovation (62). Proactively addressing these inequities is a critical prerequisite for the responsible and effective future deployment of AI in healthcare.

 Dynamic monitoring and dis ease prediction

AI plays a crucial role in dynamic monitoring and disease prediction by continuously analyzing changes in patients’ vital signs and biomarkers to support clinical decision-making (30). Technically, RL algorithms such as Deep Q-Networks (DQN) learn adaptive alerting or treatment policies from longitudinal vital-sign and laboratory data streams by optimizing long-term rewards (63, 64).

A compelling application of this approach is demonstrated in cardiovascular disease management. Hinrichs et al. (65), for instance, conducted an analysis of data from the TIM-HF2 trial, which incorporated daily transmissions from non-invasive monitoring devices. They developed a machine learning model to predict the risk of heart failure hospitalization within seven days. The AI model reliably identified high-risk patients requiring immediate intervention, outperforming a conventional heuristic algorithm in terms of the area under the ROC curve and significantly reducing the need for manual data review.

A comparative analysis, however, highlights both the promise and the persisting challenges. While AI demonstrates significant potential in dynamic forecasting, its effectiveness is constrained by several factors. Firstly, model performance hinges on high-quality, real-time data streams and carefully specified reward signals (63). Secondly, akin to many complex AI systems, the transparency and interpretability of the model’s decision-making process remain a major barrier to its widespread adoption and trust in clinical practice (44). Addressing these twin challenges of data quality and model interpretability is pivotal for advancing the field toward mature clinical application.

In summary, AI applications are revolutionizing disease diagnosis. The transition from precise diagnostic insights to personalized therapeutic actions forms the foundation of a closed-loop precision medicine system, which is the focus of the next section on treatment management.

 Treatment management in a closed loop

This section explores how AI leverages diagnostic insights to formulate, execute, and dynamically adjust personalized treatment strategies, thereby establishing an optimized closed-loop system for clinical intervention.

Personalized treatment plans

In precision medicine, imaging CNNs, genomic Random Forests, and RL/DQN policies are increasingly integrated for sequential decision-making and therapy adaptation (50, 51, 63, 64).

AI’s application in personalized treatment plan formulation is particularly prominent, as it leverages ML technologies to identify key biomarkers from a patient’s genetic information, predict drug responses, and provide individualized treatment decision support for clinicians (66). Multiple studies show that AI models have broad potential in oncology. For example, Rakaee et al. (67) developed and evaluated a DL model to predict the response of advanced non-small cell lung cancer (NSCLC) patients to immune checkpoint inhibitors (ICI), achieving high prediction accuracy. Similarly, Du et al. (68) developed an MRI-based self-attention network (MESN) to predict pathological complete response (pCR) in breast cancer patients, which outperformed traditional models (69). IBM Watson for Oncology is a notable example, providing evidence-based recommendations based on a patient’s specific molecular profile (70).

Despite its potential, ensuring the generalizability and validation of these models across diverse populations remains a critical challenge.

 Drug dosing optimization

Pharmacogenomic ML predicts patient response to drugs, while RL-style control adjusts dosing in real-time using streaming data from labs and vital signs (63, 64, 71).

Building on personalized treatment plans, AI plays a key role in the individualization of drug dosing. By analyzing a patient’s physiological characteristics and genetic background, AI can predict the rate of drug metabolism and efficacy, allowing for the development of personalized drug dosages that improve efficacy and reduce side effects (72). For example, AI has been used in genetic testing for warfarin dosing to predict optimal doses based on individual genetic markers (73). Furthermore, AI systems have the capability to dynamically adjust treatment plans, continuously monitoring patients’ treatment responses and modifying plans accordingly. This ability is especially critical in the management of chronic diseases (74). Technologies such as AI-driven telemedicine platforms, which enable real-time monitoring and management, are increasingly being integrated into healthcare settings (75). Nevertheless, challenges related to data privacy, real-time data quality, and system interoperability must be overcome. Concerns regarding the accessibility and applicability of genetic testing in resource-limited settings also remain a significant barrier.

 AI-assisted surgery and robotics

In robotic and image-guided procedures, computer-vision DL models provide perception (segmentation, tracking) and decision support for guidance and safety (6, 30).

Beyond drug dosing and treatment planning, emerging technologies such as AI-assisted surgery and robotics are transforming treatment management. For example, Intuitive Surgical’s da Vinci system, a robotic surgery platform powered by AI, enables more precise and minimally invasive procedures (76). In neuro-rehabilitation, AI systems are being used to personalize rehabilitation programs, adjusting therapeutic interventions based on real-time patient data, improving outcomes in stroke recovery (77). On the other hand, these technologies require further development to overcome regulatory and technical challenges.

AI, through deep analysis of individual data, aids in creating personalized treatment plans, optimizing therapeutic effects, and reducing side effects. Nevertheless, challenges related to data integration, privacy, and the development of universally applicable AI models need to be addressed.

 Innovation in drug development

This section examines how AI is accelerating and optimizing various stages of the drug development pipeline, from initial discovery to clinical deployment.

Drug discovery and personalized design

DL models learn structure–activity relationships and protein/ligand representations, while GANs augment scarce assay or imaging data to improve robustness during candidate discovery (78–81).

AI occupies a central role in drug development. By leveraging DL and ML technologies, AI can analyze vast compound databases, identify potential drug candidates, and predict their activity and toxicity, significantly accelerating the drug screening process (82). For example, AlphaFold, a breakthrough AI model, has provided unprecedented insights into protein structures, which are essential for drug target identification (83). Additionally, by combining patient genetic and biomarker information, AI assists in the design of personalized treatment plans, improving drug efficacy and reducing adverse reactions (84). IBM Watson for Drug Discovery is a notable example, integrating patient data to accelerate the development of personalized cancer therapies (85).

However, challenges such as the reliability of AI predictions in real-world populations require further validation. The integration of multi-modal data (genetic, clinical, environmental) is still in its infancy, and data privacy and ethical concerns remain major barriers.

 Clinical trial optimization

NLP (e.g., BERT) parses eligibility criteria and outcomes from protocols and EHRs, while RL supports adaptive randomization and simulation for trial optimization (57, 64).

Building on the advancements in drug discovery, AI contributes to optimizing clinical trial design by precisely selecting trial participants, improving trial efficiency, and predicting trial results. For instance, Manz et al. (86) developed an AI-powered clinical trial simulation tool to predict outcomes and improve recruitment strategies, reducing trial timelines and ensuring that participants most likely to benefit from the treatment are included (87). A key challenge is the generalizability of AI models across different demographics and regions, which needs further exploration.

 Drug repositioning and interaction prediction

Literature - and omics - driven repositioning blends clinical NLP with ML representations to propose new indications for existing drugs (56, 57). Similarly, clinical NLP and sequence/graph ML extract medication entities and relations from EHRs and biomedical text to surface plausible drug–drug interactions (56, 57).

Beyond new drug development, AI shows remarkable advantages in drug repositioning, enabling the discovery of new indications for existing drugs, thereby offering patients more treatment options (88). For example, Deep Repositioning has been used to identify novel applications of approved drugs for rare diseases (89). In addition, AI can predict drug interactions, assisting clinicians in developing safe and rational drug regimens, particularly for patients requiring polypharmacy (74). By analyzing large clinical databases, AI can detect potential interactions. IBM Watson for Oncology has been extended to predict drug interactions in cancer treatment regimens, with a recent study demonstrating enhanced capability by integrating a knowledge graph, building upon its core purpose of providing evidence-based recommendations and aligning with data-driven drug safety profiling (70).

Nevertheless, there is still a need for validation in clinical settings for both repositioned drugs and predicted interactions, as the potential for side effects must be carefully considered. AI systems must be further developed to incorporate real-world clinical data and provide actionable, clinically relevant results.

 Emerging technologies in drug development

Generative models (e.g., GANs) and reinforcement learning can simulate patient trajectories and augment sparse datasets to inform digital-twin-style exploration and delivery strategies (64, 78, 80).

Looking to the future, emerging technologies such as digital twins are increasingly shaping drug development. Digital twins, which create virtual representations of patients, can simulate how a patient might respond to different drugs, offering real-time feedback for drug development and personalized treatment (90). Similarly, AI-assisted robotics enhance the precision of drug delivery and treatment, particularly in fields like orthopedics and neurology (91–93).

That said, integrating these technologies into drug development pipelines requires overcoming significant regulatory and technical challenges.

Overall, AI enhances the drug development process across multiple stages, effectively improving development efficiency and success rates while providing more accurate treatment options for patients. However, challenges related to data validation, ethical considerations, and the integration of emerging technologies need to be addressed.

 Full-cycle health monitoring

This section outlines the role of AI and connected devices in enabling continuous, personalized health management outside traditional clinical settings.

Real-time monitoring with wearable devices

On-device DNNs denoise and compress biosignals, while RL/DQN and time-series models track longitudinal risk and recommend just-in-time actions (6, 63, 64, 94).

AI, combined with wearable devices (e.g., smartwatches, biosensors), shows immense potential in health management. These devices continuously monitor users’ physiological parameters, creating individualized health profiles. AI analyzes these data to assess potential health risks, such as evaluating cardiovascular health status through heart rate variability (95). A notable example is the Heart Study, which employs AI algorithms to identify cardiac arrhythmias through smartwatch-derived data (96).

Nevertheless, challenges remain in data standardization, data privacy, and the accuracy of AI predictions in diverse populations.

 Disease prediction and early warning

Sequential DL and RL frameworks forecast short-term physiological trends from continuous data streams (63, 64).

Building on real-time monitoring, AI technology uses historical data and ML algorithms to predict disease onset trends. For diabetic patients, AI can predict blood sugar fluctuations based on blood sugar levels, diet, and activity, issuing early warnings and guiding preventive measures (97). Diabetes Watch is a notable example, analyzing real-time data to provide insights and recommendations for managing diabetes (98). On the other hand, issues related to data quality and the need for large, diverse datasets for accurate predictions persist.

 Real-time alerts and personalized management

Signal-processing with CNNs plus policy learning enables real-time anomaly detection and triage from wearable and bedside monitors (6, 64). NLP-driven coaching systems and RL personalize lifestyle recommendations and education using longitudinal data (58, 64, 99).

AI achieves real-time monitoring of patients’ health status, automatically pushing warnings or interventions when necessary (e.g., for abnormal heart rate patterns) (100). Research on wearable biosensors indicates promising approaches for non-invasive biomarker detection (101). Moreover, AI facilitates personalized health management, offering tailored lifestyle recommendations (e.g., personalized exercise and diet plans) based on individual health data (102). The Cardiogram app, for instance, uses AI to analyze heart rate data and offers personalized lifestyle recommendations (103).

However, issues such as data accuracy, system integration with existing healthcare infrastructures, and the management of false positives remain critical. Concerns about the accessibility of wearable devices and ensuring equitable access to AI-powered health management solutions, especially in low-resource settings, also remain.

 Integration with telemedicine platforms

Multimodal models fuse wearable streams with clinical text and images to summarize patient state for remote decision support (13, 41). By embedding these fusion models into telemedicine backends, clinicians can receive push-style, evidence-graded summaries without manual data review.

The practical value of this integration is demonstrated by research into AI-powered remote monitoring systems. These systems synthesize real-time data from wearable devices (e.g., for monitoring heart failure or chronic obstructive pulmonary disease) with patients’ EHRs (65, 95). By applying predictive analytics, they can identify patients at high risk of clinical deterioration, enabling timely intervention by healthcare providers. Studies have shown that such integrated approaches can reduce hospital readmission rates for patients with chronic conditions by facilitating proactive and personalized care management (65, 95).

The advancement of these integrated platforms, however, underscores persistent challenges. Key issues that must be addressed include achieving robust cross-platform data interoperability, ensuring the security of longitudinal data across the care continuum, and conducting rigorous validation of AI-driven clinical summaries across diverse healthcare settings and patient populations to ensure their reliability and generalizability.

 Paradigm innovation: AI-driven transformation of the healthcare system

AI applications in precision medicine and nursing are driving a significant transformation in healthcare systems, transitioning them from reactive to proactive care. This shift involves not only personalized healthcare, remote monitoring, and intelligent decision support systems but also improvements in hospital logistics, patient flow, and administrative efficiency. Figure 4 illustrates the operational flow of AI algorithms in clinical settings, showing how patient data is processed to provide diagnostic and treatment recommendations, and how feedback loops optimize AI models. This figure highlights the key stages of AI integration, from data input to decision support and model optimization, providing a comprehensive view of AI’s role in reshaping healthcare workflows. The following subsections delve into the specific ways AI is transforming service models, system capabilities, and the healthcare ecosystem.

AI algorithm operational flow in clinical settings. This figure illustrates the workflow from data input (e.g., EHRs, medical imaging) to decision support and feedback loops, highlighting the integration of AI technologies in clinical settings.

Flowchart depicting an AI-driven healthcare process. Stages: Data Input (collect patient data), Data Preprocessing (clean and normalize data), AI Processing (train and execute models), Decision Support (diagnostic and treatment recommendations), and Feedback Loop (validate and optimize models). Each stage involves specific tasks aimed at improving data-driven healthcare decision-making. Service model reconstruction

Personalized medicine is a key component of AI-driven transformation. AI tailors treatment plans by integrating individual genetic information, biomarkers, and lifestyle factors, enabling more effective and customized treatments. For example, AI systems have been used in oncology to analyze genomic data, identifying specific tumor mutations and helping clinicians select the most effective targeted therapies for cancer patients (104). Additionally, AI can predict patient responses to drugs and optimize dosage settings to minimize adverse reactions (105). AI-driven tools such as IBM Watson for Oncology use AI to analyze patient data, including genetic profiles, to suggest personalized treatment options based on the latest research and clinical data, significantly improving the speed and accuracy of treatment plans (106).

Moreover, AI is being used to predict and monitor chronic conditions, leading to a shift from traditional reactive care to proactive, preventive care. For instance, AI plays a role in managing diabetes by predicting blood sugar fluctuations and adjusting insulin doses automatically, thereby reducing the risk of emergencies and improving long-term patient health (107).

 System capability upgrades

AI has fundamentally altered healthcare through the development of remote monitoring and predictive analytics, greatly enhancing both the efficiency of patient care and the allocation of medical resources. For example, AI-powered systems such as remote ECG monitoring platforms use wearable devices to collect patient data in real-time. These systems can analyze heart rate, rhythm, and activity levels to predict potential heart attacks or strokes before they occur (108). In practical applications, Li et al.’s (109) research shows that wearable devices integrated with AI algorithms continuously monitor patients and alert clinicians to early signs of deterioration, reducing hospital readmission rates and minimizing unnecessary emergency visits.

Furthermore, AI’s integration with IoT devices supports seamless, dynamic health management, particularly for older adult patients with chronic conditions (110). AI models predict health deterioration in real time, enabling timely intervention and reducing hospital congestion (111). By enabling continuous health assessments, AI in remote monitoring not only improves the patient’s quality of life but also reduces the strain on hospitals and clinics, particularly in managing large patient populations in resource-limited settings (112).

 Ecosystem evolution

Intelligent Decision Support Systems (IDSS) are increasingly used to process vast amounts of medical data, helping healthcare professionals make faster and more accurate decisions (113). For example, studies have shown that AI models deployed in radiology departments assist clinicians by identifying patterns in medical images such as CT scans and MRIs, thereby improving diagnostic accuracy and treatment outcomes (114). These systems are also integrated with EHR, enabling personalized treatment recommendations based on patient characteristics and medical history (114).

AI-driven systems are optimizing hospital logistics and patient flow by predicting patient demand, improving resource allocation, and reducing wait times. For instance, AI algorithms predict emergency department admission rates, helping hospital managers allocate resources more efficiently and avoid bottlenecks (115). This proactive management of patient flow minimizes overcrowding in hospitals, reduces patient waiting times, and optimizes bed usage, leading to better patient experiences and improved operational efficiency (116).

Additionally, AI is significantly enhancing administrative efficiency by automating routine tasks such as scheduling, billing, and documentation (117, 118). For example, AI-powered administrative assistants are now helping hospitals streamline patient scheduling by predicting the most optimal times for appointments and automatically adjusting to cancelations or no-shows, thereby reducing administrative overhead and improving service availability (119).

In conclusion, the integration of AI in precision medicine, remote monitoring, and intelligent decision support systems is driving a shift from reactive to proactive care in healthcare. Real-world applications such as AI-driven personalized cancer treatments, predictive health monitoring using IoT, and automated administrative processes are fundamentally transforming how healthcare is delivered. AI is enhancing hospital logistics, patient flow, and resource allocation, leading to improved hospital operations and greater overall efficiency. As technology continues to evolve, AI will play an increasingly pivotal role in providing precise, proactive, and personalized healthcare experiences, optimizing both clinical outcomes and operational efficiency.

 Development challenges and future prospects

Table 3 (Summary of Development Challenges and Future Prospects in AI Healthcare) outlines the key ethical, privacy, and governance challenges in AI healthcare implementation, along with their impacts and potential future directions. Addressing these issues is crucial for ensuring the responsible and equitable deployment of AI technologies in healthcare. By integrating robust ethical frameworks, enhancing algorithmic transparency, and fostering cross-institutional collaboration, we can mitigate the risks associated with AI deployment while maximizing its potential to transform healthcare delivery. The future of AI in healthcare lies in its ability to align with human values, ensuring equitable access and trust among stakeholders.

Summary of development challenges and future prospects in AI healthcare.

Domain Key challenges Impact Future directions
Ethical challenges Data privacy, algorithmic transparency, fairness, and bias in AI models Hinders trust, equitable healthcare delivery, and adoption of AI systems Implement XAI, federated learning, and fairness protocols; ensure diverse datasets
Privacy constraints Data sharing barriers, regulatory compliance (e.g., HIPAA, GDPR), and data leaks Limits AI development and cross-institutional collaboration Use secure data-sharing methods (e.g., blockchain) and federated learning
Governance and accountability Lack of clear liability frameworks for AI errors Delays AI adoption and raises ethical concerns Develop standardized governance structures and define roles for AI developers and practitioners
Algorithmic bias Biased datasets leading to unequal outcomes for diverse populations Perpetuates healthcare disparities and reduces AI reliability Use data diversification, fairness auditing, and algorithmic transparency tools
Clinical limitations AI tools may fail to generalize in real-world settings; over-reliance on AI Erosion of clinical skills and potential patient harm Continuous validation, human oversight, and training for healthcare professionals
Regulatory challenges Varying regional regulations, long approval timelines, and lack of standardization Hinders multinational AI adoption and integration Develop international AI regulatory frameworks and clear certification guidelines
Institutional resistance Concerns about job displacement, lack of infrastructure, and trust in AI Slows AI adoption and integration into healthcare systems Invest in training programs, infrastructure, and foster trust in AI’s role in care
Symbiotic AI (SAI) Not applicable Not applicable Develop tools integrating SAl, using XAl and FL to advance healthcare Al

XAI, Explainable Artificial Intelligence; HIPAA, Health Insurance Portability and Accountability Act; GDPR, General Data Protection Regulation; SAI, Symbiotic Artificial Intelligence; FL, federated learning.

 Heterogeneity of clinical evidence and generalizability challenges

A critical synthesis of the literature presented in this review reveals significant disparities in the maturity of clinical evidence supporting AI applications across different medical specialties. In well-established domains such as oncology (e.g., breast cancer screening with mammography) and cardiology (e.g., echocardiogram interpretation), AI models have been validated in larger, often multi-center studies. Crucially, a growing body of comparative studies and meta-analyses in these fields now provides evidence for benchmarking AI performance directly against conventional standard care. For specific, narrow tasks, some deep learning models demonstrate performance on par with or even surpassing that of human experts, with reported metrics frequently achieving AUROC ranges of 0.89 to 0.90 (94, 120–122). However, it is essential to note that this technical equivalence or superiority in controlled studies does not automatically translate to superior patient outcomes in routine practice, and evidence demonstrating a clear improvement in final patient endpoints over standard care remains less common.

Conversely, as this review identifies, areas such as infectious disease, maternal-fetal medicine, and pediatrics remain relatively nascent. The evidence here is often derived from smaller, single-center, retrospective cohorts, making direct comparisons with the efficacy of established standard care approaches difficult and highlighting an urgent need for more robust external validation and prospective trials (123).

Beyond the variability in evidence across specialties, a paramount and cross-cutting challenge is the generalizability of AI models between high-resource and low-resource settings. Models excelling in data-rich environments often suffer from performance degradation due to “domain shift” when faced with different patient demographics, imaging equipment, or clinical practices in resource-constrained settings (124). For instance, a landmark comparative study by Zech et al. (47) demonstrated that a chest X-ray model trained on data from one US hospital system experienced a significant drop in performance when applied to data from other hospitals within the same network, thereby failing to generalize its initially promising performance versus local diagnostic standards. This underscores that an AI model’s performance cannot be evaluated in isolation but must be proven robust and superior to relevant local standard care across diverse environments.

Addressing these challenges requires a dual approach: first, the promotion of rigorous, multi-center clinical trials designed specifically to compare AI-integrated care pathways against current standard of care and the synthesis of existing evidence through systematic reviews and meta-analyses to establish robust performance benchmarks; and second, the technical development and adoption of strategies like federated learning for privacy-preserving multi-institutional collaboration and domain adaptation methods to enhance model robustness across diverse clinical environments.

 Ethical challenges in AI implementation

The deployment of AI in healthcare brings forth significant ethical and privacy constraints that can impede its widespread use. Ethical concerns regarding data privacy, algorithm transparency, and fairness must be carefully addressed to build trust and ensure equitable healthcare delivery. Data privacy remains one of the most significant ethical challenges in AI implementation. Healthcare data, including sensitive patient information, must be handled with the utmost care to protect privacy and comply with regulations such as the Health Insurance Portability and Accountability Act (HIPAA) in the United States or the General Data Protection Regulation (GDPR) in Europe (125, 126). However, despite the benefits of data sharing in AI development, these regulations often create friction, as sharing high-quality data across institutions becomes complicated due to privacy concerns (127). Even though solutions such as FL can mitigate privacy risks by allowing data to stay localized, they are still not sufficient in addressing the real-world complexities of cross-institutional collaboration (128). Moreover, the challenge of ensuring secure data transmission and preventing breaches remains a constant concern, as healthcare organizations continue to grapple with high-profile data leaks (129).

Another key ethical issue in AI deployment is the lack of algorithmic transparency. AI models are often considered “black boxes” in healthcare, meaning their decision-making processes are not easily understood by healthcare professionals (130). This lack of interpretability raises concerns about the trustworthiness of AI systems in critical healthcare decisions. If healthcare professionals cannot fully understand the rationale behind AI recommendations, they may hesitate to rely on them, potentially delaying or undermining patient care (131, 132). To address these concerns, it is essential that future AI models incorporate XAI techniques such as knowledge graphs, causal inference, and visual analytics (133, 134). These methods will not only help clarify the decision-making process but also foster collaboration between AI systems and healthcare professionals, ensuring that AI is viewed as a tool to support—not replace—human judgment.

Furthermore, the issue of fairness in AI remains a crucial challenge. AI systems are often trained on datasets that lack diversity, resulting in biased models that may not accurately reflect the needs of underrepresented or vulnerable populations. For instance, Kiyasseh et al. (135) found that AI model could reliably assess surgical performance but exhibited bias. AI models trained primarily on data from affluent, homogeneous populations may not perform as well for patients from different racial, ethnic, or socioeconomic backgrounds, this raises serious concerns about equity in healthcare, as AI systems could inadvertently perpetuate disparities in diagnosis, treatment, and patient outcomes (136). For instance, Obermeyer et al. (54) found that an AI tool used in US hospitals to predict additional medical care disproportionately favored White patients over Black patients. Correcting this bias increased the percentage of Black patients receiving additional care from 17.7 to 46.5%, demonstrating the significant impact of algorithmic bias in clinical outcomes. To mitigate these biases, it is crucial to ensure that AI systems are designed to be fair and inclusive, with particular attention paid to the representation of minority groups in training datasets. Recent studies have proposed solutions such as algorithmic auditing tools, and collaborations between AI developers and ethicists to establish clear AI fairness protocols (137–139).

 Privacy and ethical constraints in real-world healthcare

In practical terms, privacy concerns and ethical constraints can significantly impede the adoption of AI in healthcare. Healthcare institutions often face barriers related to data sharing due to patient privacy rights, institutional reluctance, and regulatory challenges (140). Even when data-sharing technologies such as FL or blockchain are employed, the complexity of synchronization and data sharing remains a significant hurdle (141). This creates challenges in developing AI models that require vast and varied datasets to enhance accuracy and ensure generalizability across diverse populations and clinical practices.

Moreover, ethical dilemmas arise when algorithms are used to make decisions that impact patient care without clear accountability. For example, if an AI system makes an incorrect diagnosis that leads to patient harm, determining responsibility becomes problematic. This issue is further complicated by the lack of clear governance frameworks for AI in healthcare. Currently, most healthcare systems do not have well-established processes to determine liability when AI systems make errors, which may hinder the widespread adoption of AI technologies (142, 143). Addressing these issues requires the development of clear, standardized governance structures that define the roles and responsibilities of healthcare practitioners, AI developers, and institutions (144–146).

Furthermore, addressing algorithmic bias is paramount to ensuring equitable AI performance across diverse patient populations. This requires the implementation of validated bias mitigation strategies—such as reweighting and adversarial debiasing during model training—and the rigorous application of fairness metrics (e.g., demographic parity, equalized odds) to quantify and monitor performance disparities across different demographic groups (147). Proactively auditing models for these metrics is essential to prevent the exacerbation of existing health disparities.

In addition to governance concerns, healthcare institutions may also face institutional resistance to adopting AI technologies. This resistance may stem from concerns about job displacement, lack of technological infrastructure, or trust in AI systems (148). To facilitate AI adoption, training programs for healthcare professionals are essential, especially to address concerns about the erosion of clinical skills when AI is over-relied upon (149). Additionally, healthcare providers must be educated on how AI can enhance rather than replace their clinical expertise, fostering trust in AI’s role in improving patient care (150). Institutional commitment to AI integration is key, and this requires substantial investments in both training and infrastructure development (151).

 Limitations and unintended consequences of AI in healthcare

While AI holds immense potential to transform healthcare, it is essential to acknowledge its limitations and the unintended consequences that may arise from its deployment. Algorithmic bias, as discussed in Section 5.1, remains a significant challenge in clinical settings. AI systems trained on biased datasets can lead to unequal outcomes for different demographic groups, such as patients from diverse racial, ethnic, or socioeconomic backgrounds. Recent studies have proposed methods such as data diversification and fairness auditing to mitigate these biases (152, 153).

In clinical practice, AI-driven diagnostic tools may fail to generalize in real-world settings due to variations in patient populations, data quality, and clinical workflows (154). Furthermore, over-reliance on AI systems can result in the erosion of clinical skills among healthcare professionals, who may become overly dependent on AI recommendations (155). AI’s failure to account for real-world complexities also highlights the need for continuous validation and human oversight in healthcare AI deployment.

Regulatory challenges further complicate the integration of AI in healthcare. Different regions have varying regulations governing AI use in clinical settings, creating complexities for multinational healthcare systems. Issues such as the lack of standardization, long approval timelines, and concerns over data privacy and security hinder the widespread adoption of AI technologies in clinical practices (156). Researchers have proposed solutions such as the development of international AI regulatory frameworks and the establishment of clear guidelines for AI certification and accountability (157, 158).

Ethical concerns, particularly regarding data privacy and patient consent, must also be addressed. AI systems often require access to sensitive patient data, and it is crucial to ensure that patient consent is obtained transparently and that their privacy is safeguarded (159, 160). The implementation of XAI is one approach to improving transparency and trust, as it can help healthcare professionals better understand AI-driven decisions and incorporate them into their clinical practice (161).

Finally, there are practical barriers to AI adoption, including insufficient technological infrastructure, lack of training among healthcare practitioners, and institutional resistance to adopting AI technologies (162). Many healthcare systems, particularly in low-resource settings, may not have the necessary infrastructure to support AI tools, such as high-performance computing or sufficient data storage capacity (62). In addition, healthcare practitioners need appropriate training to effectively use AI systems, and they must trust AI’s potential to improve patient care (163). Overcoming these barriers requires both institutional commitment to technology adoption and comprehensive training programs for clinicians.

 Future directions in ethical AI implementation

Looking forward, the development of robust ethical governance frameworks is essential to address the multifaceted challenges posed by AI in healthcare. These frameworks must prioritize patient consent, data security, algorithmic transparency, and fairness to build trust and ensure equitable healthcare delivery. Institutions should adopt policies that not only harness the benefits of AI but also mitigate potential risks, ensuring that AI technologies are deployed responsibly and ethically (164). A comprehensive approach is essential, encompassing secure data-sharing methods, enhanced transparency in algorithmic decision-making, and the establishment of clear ethical guidelines for AI deployment in healthcare (165).

Recent international initiatives have underscored the importance of coherent global AI governance frameworks in healthcare. The U.S. Food and Drug Administration (FDA) has proposed the Total Product Lifecycle (TPLC) regulatory approach for AI/ML-based medical devices, emphasizing continuous monitoring and adaptive learning within a controlled framework (166). Similarly, the European Union’s AI Act classifies AI systems based on risk categories, requiring transparency, human oversight, and post-market audits for high-risk applications in healthcare (167). The World Health Organization (WHO) has also introduced its “Ethics and Governance of Artificial Intelligence for Health” guidance, outlining six key principles—autonomy, safety, transparency, accountability, inclusiveness, and sustainability—to guide ethical AI use globally (168). Incorporating these frameworks can promote international harmonization of AI governance and ensure consistent accountability across healthcare systems.

In addition to global regulation, enhancing data accountability and clinical oversight remains critical. Healthcare institutions should adopt auditable data governance mechanisms, including version-controlled data pipelines, algorithmic logging, and external audit trails to ensure traceability of AI-driven decisions (133, 169). These mechanisms are the foundation for the practical, long-term maintenance of AI systems, enabling continuous monitoring for model drift—such as performance degradation due to shifts in clinical practice or patient demographics—and triggering the need for timely model recalibration or retraining (170). Transparent auditing not only improves data integrity but also fosters public trust. Furthermore, clinician oversight should be institutionalized by establishing AI oversight committees that review and validate high-impact AI recommendations, ensuring that human judgment remains central to clinical decision-making (171).

The emerging framework of Symbiotic AI (SAI) provides a promising approach to ethical AI implementation. SAI emphasizes a collaborative relationship between AI systems and human users, fostering mutual understanding and trust. In healthcare, SAI addresses critical ethical challenges such as data privacy, algorithmic transparency, and fairness by adopting a human-centered approach to AI design (172). For instance, SAI frameworks integrate XAI techniques to enhance the interpretability of AI decision-making for healthcare professionals, thereby mitigating the “black box” problem (173). These approaches make AI decisions more understandable and transparent, facilitating better integration into clinical workflows.

To clarify its conceptual boundaries, SAI should be distinguished from related frameworks such as Explainable AI (XAI) and human–AI teaming. While XAI primarily focuses on interpretability and transparency, and human–AI teaming emphasizes cooperative task execution, SAI extends beyond both by promoting continuous bidirectional learning between humans and AI systems (174–176). In operational terms, SAI functions as a mutually adaptive system: clinicians refine their decisions based on AI insights, while AI models dynamically learn from clinician feedback and contextual data (175, 177). Critically, this feedback loop should be expanded to include patients, ensuring that their experiences, preferences, and reported outcomes inform both clinical use and the iterative refinement of AI tools. This iterative process ensures that AI remains aligned with ethical standards, clinical reasoning, patient values, and human values.

Existing studies have demonstrated both successes and challenges in the deployment of SAI systems in healthcare. For example, Tschandl et al. (178) found that SAI frameworks incorporating XAI techniques significantly improved clinician trust and understanding of AI-driven diagnostic recommendations. However, Mollura et al. (179) reported difficulties in scaling SAI systems due to computational cost and the need for extensive data storage, especially in low-resource settings. In such environments, the computational demands of SAI models could limit their accessibility and effectiveness.

Furthermore, SAI promotes the use of FL and secure data-sharing protocols, ensuring patient privacy while facilitating cross-institutional collaboration (180). FL allows for collaborative model training without sharing sensitive patient data, which is particularly valuable in maintaining privacy across healthcare institutions with varying data access policies.

To address these challenges, lightweight AI alternatives are being explored to reduce the computational burden. For example, model pruning, quantization, and edge computing offer potential solutions for implementing SAI in low-resource environments (181). These lightweight models can achieve performance comparable to more computationally intensive models while reducing the need for high-performance computing infrastructure.

By focusing on the symbiotic relationship between AI and humans, SAI provides a robust framework for ethical AI deployment in healthcare. AI systems must be designed to align with human values and societal needs, ensuring that technological advancements contribute to ethical and equitable healthcare solutions.

 Conclusion and implications for public health

This narrative review has synthesized evidence demonstrating the transformative role of AI across the precision medicine continuum, from diagnosis and treatment personalization to drug discovery and continuous health monitoring. These advancements signal a pivotal shift toward more proactive, data-driven healthcare.

However, the translation of this potential into widespread clinical practice is contingent upon addressing significant challenges. These include the need for more robust, multi-center clinical validation to solidify the evidence base, and the critical imperative to overcome issues of data privacy, algorithmic fairness, and - most importantly - model generalizability across diverse populations and healthcare settings. Tackling these challenges is essential to ensure that the benefits of AI do not exacerbate existing health disparities but rather contribute to greater health equity.

The integration of AI into precision medicine thus holds profound implications for public health. AI can empower a shift from reactive treatment to population-wide preventive health by enabling targeted screening and early intervention strategies, potentially reducing the long-term burden of disease. To realize this vision, future efforts must focus on developing AI models with inherent fairness and robustness, fostered through diverse data collection and collaborative frameworks like federated learning. Concurrently, the establishment of supportive policy and ethical guidelines is crucial to steward the equitable and trustworthy deployment of these powerful technologies, ensuring they serve the goal of improving health for all.

 Author contributions

QZ: Methodology, Conceptualization, Writing – review & editing, Writing – original draft, Investigation. CH: Formal analysis, Validation, Data curation, Writing – review & editing. JZ: Project administration, Supervision, Writing – review & editing.

 Acknowledgments

The authors would like to thank all participants for participating in this clinical trial.

 Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

 Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

 Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

 References RamaswamiR BayerR GaleaS. Precision Medicine from a Public Health Perspective. Annu Rev Public Health. (2018) 39:153–68. doi: 10.1146/annurev-publhealth-040617-014158, 29166244 YuK-H BeamAL KohaneIS. Artificial intelligence in healthcare. Nature Biomed Eng. (2018) 2:719–31. doi: 10.1038/s41551-018-0305-z, 31015651 Ehteshami BejnordiB. VetaM. Johannes DiestP.van GinnekenB. KarssemeijerN. 2017 Diagnostic assessment of deep learning algorithms for detection of lymph node metastases in women with breast cancer JAMA 318 2199–2210 doi: 10.1001/jama.2017.14585, 29234806 XieW LiY MengX ZhaoM. Machine learning prediction models and nomogram to predict the risk of in-hospital death for severe DKA: A clinical study based on MIMIC-IV, eICU databases, and a college hospital ICU. Int J Med Inform. (2023) 174:105049. doi: 10.1016/j.ijmedinf.2023.105049, 37001474 SavovaGK DanciuI AlamudunF MillerT LinC BittermanDS . Use of natural language processing to extract clinical cancer phenotypes from electronic medical records. Cancer Res. (2019) 79:5463–70. doi: 10.1158/0008-5472.Can-19-0579, 31395609 EstevaA KuprelB NovoaRA KoJ SwetterSM BlauHM . Dermatologist-level classification of skin cancer with deep neural networks. Nature. (2017) 542:115–8. doi: 10.1038/nature21056, 28117445 HuangSC ChenCC LanJ HsiehTY ChuangHC ChienMY . Deep neural network trained on gigapixel images improves lymph node metastasis detection in clinical settings. Nat Commun. (2022) 13:3347. doi: 10.1038/s41467-022-30746-1, 35688834 WangG LiuX YingZ YangG ChenZ LiuZ . Optimized glycemic control of type 2 diabetes with reinforcement learning: a proof-of-concept trial. Nat Med. (2023) 29:2633–42. doi: 10.1038/s41591-023-02552-9, 37710000 ZhangK YangX WangY YuY HuangN LiG . Artificial intelligence in drug development. Nat Med. (2025) 31:45–59. doi: 10.1038/s41591-024-03434-4, 39833407 DeviDH DuraisamyK ArmghanA AlsharariM AliqabK SorathiyaV . 5G technology in healthcare and wearable devices: a review. Sensors. (2023) 23:519. doi: 10.3390/s23052519, 36904721 Olivares-GandyHJ Domínguez-IsidroS López-DomínguezE Hernández-VelázquezY Tapia-McClungH Jorged-l-C. A telemonitoring system for nutritional intake in patients with chronic kidney disease receiving peritoneal dialysis therapy. Comput Biol Med. (2019) 109:012. doi: 10.1016/j.compbiomed.2019.04.012 WalkdenJ-A McCullaghPJ KernohanWG. Patient and carer survey of remote vital sign telemonitoring for self-management of long-term conditions. BMJ Health & Care Informatics. (2019) 26:79. doi: 10.1136/bmjhci-2019-100079, 31857341 MoorM BanerjeeO AbadZSH KrumholzHM LeskovecJ TopolEJ . Foundation models for generalist medical artificial intelligence. Nature. (2023) 616:259–65. doi: 10.1038/s41586-023-05881-4, 37045921 NgZQP LingLYJ ChewHSJ LauY. The role of artificial intelligence in enhancing clinical nursing care: a scoping review. J Nurs Manag. (2022) 30:3654–74. doi: 10.1111/jonm.13425, 34272911 TeoZL JinL LiS MiaoD ZhangX NgWY . Federated machine learning in healthcare: a systematic review on clinical applications and technical architecture. Cell Reports Med. (2024) 5:101419. doi: 10.1016/j.xcrm.2024.101419, 38340728 AungYYM WongDCS TingDSW. The promise of artificial intelligence: a review of the opportunities and challenges of artificial intelligence in healthcare. Br Med Bull. (2021) 139:4–15. doi: 10.1093/bmb/ldab016, 34405854 LibbrechtMW NobleWS. Machine learning applications in genetics and genomics. Nat Rev Genet. (2015) 16:321–32. doi: 10.1038/nrg3920, 25948244 Milne-IvesM de CockC LimE ShehadehMH de PenningtonN MoleG . The effectiveness of artificial intelligence conversational agents in health care: systematic review. J Med Internet Res. (2020) 22:e20346. doi: 10.2196/20346, 33090118 TekkeşinAİ. Artificial Intelligence in Healthcare: Past, Present and Future. Anatol J Cardiol. (2019) 22:8–9. doi: 10.14744/AnatolJCardiol.2019.28661, 31670713 KantorK MorzyM. Machine learning and natural language processing in clinical trial eligibility criteria parsing: a scoping review. Drug Discov Today. (2024) 29:104139. doi: 10.1016/j.drudis.2024.104139, 39154773 ReelPS ReelS PearsonE TruccoE JeffersonE. Using machine learning approaches for multi-omics data analysis: A review. Biotechnol Adv. (2021) 49:107739. doi: 10.1016/j.biotechadv.2021.107739, 33794304 ZhaoS SuC LuZ WangF. Recent advances in biomedical literature mining. Brief Bioinform. (2021) 22:57. doi: 10.1093/bib/bbaa057, 32422651 MienyeID ObaidoG JereN MienyeE ArulebaK EmmanuelID . A survey of explainable artificial intelligence in healthcare: concepts, applications, and challenges. Inf Med Unlocked. (2024) 51:101587. doi: 10.1016/j.imu.2024.101587 YuY HeZ OuyangJ TanY ChenY GuY . Magnetic resonance imaging radiomics predicts preoperative axillary lymph node metastasis to support surgical decisions and is associated with tumor microenvironment in invasive breast cancer: a machine learning, multicenter study. EBioMedicine. (2021) 69:103460. doi: 10.1016/j.ebiom.2021.103460, 34233259 GreenhalghT ThorneS MalterudK. Time to challenge the spurious hierarchy of systematic over narrative reviews? Eur J Clin Investig. (2018) 48:e12931. doi: 10.1111/eci.12931, 29578574 RandlesR FinneganA. Guidelines for writing a systematic review. Nurse Educ Today. (2023) 125:105803. doi: 10.1016/j.nedt.2023.105803, 36996684 AvendiMR KheradvarA JafarkhaniH. A combined deep-learning and deformable-model approach to fully automatic segmentation of the left ventricle in cardiac MRI. Med Image Anal. (2016) 30:108–19. doi: 10.1016/j.media.2016.01.005, 26917105 BowrenM BrussJ ManzelK EdwardsD LiuC CorbettaM . Post-stroke outcomes predicted from multivariate lesion-behaviour and lesion network mapping. Brain. (2022) 145:1338–53. doi: 10.1093/brain/awac010, 35025994 LinM BaoG SangX WuY. Recent advanced deep learning architectures for retinal fluid segmentation on optical coherence tomography images. Sensors. (2022) 22:55. doi: 10.3390/s22083055, 35459040 RajkomarA OrenE ChenK DaiAM HajajN HardtM . Scalable and accurate deep learning with electronic health records. NPJ Digit Med. (2018) 1:18. doi: 10.1038/s41746-018-0029-1, 31304302 SmithLA Oakden-RaynerL BirdA ZengM ToM-S MukherjeeS . Machine learning and deep learning predictive models for long-term prognosis in patients with chronic obstructive pulmonary disease: a systematic review and meta-analysis. Lancet Digital Health. (2023) 5:e872–81. doi: 10.1016/S2589-7500(23)00177-2 WitowskiJ HeacockL ReigB KangSK LewinA PysarenkoK . Improving breast cancer diagnostics with deep learning for MRI. Sci Transl Med. (2022) 14:eabo4802. doi: 10.1126/scitranslmed.abo4802, 36170446 YalaA MikhaelPG StrandF LinG SmithK WanY-L . Toward robust mammography-based models for breast cancer risk. Sci Transl Med. (2021) 13:578. doi: 10.1126/scitranslmed.aba4373, 33504648 ZhangJ WuJ ZhouXS ShiF ShenD. Recent advancements in artificial intelligence for breast cancer: Image augmentation, segmentation, diagnosis, and prognosis approaches. Semin Cancer Biol. (2023) 96:11–25. doi: 10.1016/j.semcancer.2023.09.001, 37704183 AsilarE HemmerichJ EckerGF. Image based liver toxicity prediction. J Chem Inf Model. (2020) 60:1111–21. doi: 10.1021/acs.jcim.9b00713, 31978306 DimasG CholopoulouE IakovidisDK. E pluribus unum interpretable convolutional neural networks. Sci Rep. (2023) 13:11421. doi: 10.1038/s41598-023-38459-1, 37452133 KomuraD IshikawaS. Machine Learning Methods for Histopathological Image Analysis. Comput Struct Biotechnol J. (2018) 16:34–42. doi: 10.1016/j.csbj.2018.01.001, 30275936 ChenX WangX ZhangK FungK-M ThaiTC MooreK . Recent advances and clinical applications of deep learning in medical image analysis. Med Image Anal. (2022) 79:102444. doi: 10.1016/j.media.2022.102444, 35472844 TakahashiS SakaguchiY KounoN TakasawaK IshizuK AkagiY . Comparison of vision transformers and convolutional neural networks in medical image analysis: a systematic review. J Med Syst. (2024) 48:84. doi: 10.1007/s10916-024-02105-8, 39264388 ThirunavukarasuAJ TingDSJ ElangovanK GutierrezL TanTF TingDSW. Large language models in medicine. Nat Med. (2023) 29:1930–40. doi: 10.1038/s41591-023-02448-8, 37460753 YinS FuC ZhaoS LiK SunX XuT . A survey on multimodal large language models. Natl Sci Rev. (2024) 11:nwae403. doi: 10.1093/nsr/nwae403, 39679213 GulshanV PengL CoramM StumpeMC WuD NarayanaswamyA . Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. JAMA. (2016) 316:2402. doi: 10.1001/jama.2016.17216 LiptonZC. The mythos of model interpretability. Commun ACM. (2018) 61:36–43. doi: 10.1145/3233231 RudinC. Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead. Nature Machine Intelligence. (2019) 1:206–15. doi: 10.1038/s42256-019-0048-x, 35603010 ZhouSK LeHN LuuK V NguyenH AyacheN. Deep reinforcement learning in medical imaging: a literature review. Med Image Anal. (2021) 73:102193. doi: 10.1016/j.media.2021.102193, 34371440 HuangW RandhawaR JainP IczkowskiKA HuR HubbardS . Development and validation of an artificial intelligence-powered platform for prostate cancer grading and quantification. JAMA Netw Open. (2021) 4:e2132554. doi: 10.1001/jamanetworkopen.2021.32554, 34730818 ZechJR BadgeleyMA LiuM CostaAB TitanoJJ OermannEK. Variable generalization performance of a deep learning model to detect pneumonia in chest radiographs: A cross-sectional study. PLoS Med. (2018) 15:e1002683. doi: 10.1371/journal.pmed.1002683, 30399157 WynantsL Van CalsterB CollinsGS RileyRD HeinzeG SchuitE . Prediction models for diagnosis and prognosis of covid-19: systematic review and critical appraisal. BMJ. (2020) 369:m1328. doi: 10.1136/bmj.m1328, 32265220 KircherM WittenDM JainP O'RoakBJ CooperGM ShendureJ. A general framework for estimating the relative pathogenicity of human genetic variants. Nat Genet. (2014) 46:310–5. doi: 10.1038/ng.2892, 24487276 BreimanL. Random Forests. Mach Learn. (2001) 45:5–32. doi: 10.1023/A:1010933404324 ChenX IshwaranH. Random forests for genomic data analysis. Genomics. (2012) 99:323–9. doi: 10.1016/j.ygeno.2012.04.003, 22546560 Bader-El-DenM TeiteiE PerryT. Biased Random Forest For Dealing With the Class Imbalance Problem. IEEE Transactions Neural Networks Learn Syst. (2019) 30:2163–72. doi: 10.1109/TNNLS.2018.2878400, 30475733 MillerSR LuxemK LauderdaleK NambiarP HonmaPS LyKK . Machine learning reveals prominent spontaneous behavioral changes and treatment efficacy in humanized and transgenic Alzheimer's disease models. Cell Rep. (2024) 43:114870. doi: 10.1016/j.celrep.2024.114870, 39427315 ObermeyerZ PowersB VogeliC MullainathanS. Dissecting racial bias in an algorithm used to manage the health of populations. Science. (2019) 366:447–53. doi: 10.1126/science.aax2342 MartinAR KanaiM KamataniY OkadaY NealeBM DalyMJ. Clinical use of current polygenic risk scores may exacerbate health disparities. Nat Genet. (2019) 51:584–91. doi: 10.1038/s41588-019-0379-x, 30926966 AlsentzerE MurphyJR BoagW WengW-H JinD NaumannT . Publicly available clinical BERT embeddings. ArXiv. (2019) 1904:03323. doi: 10.48550/arXiv.1904.03323 LeeJ YoonW KimS KimD KimS SoCH . BioBERT: a pre-trained biomedical language representation model for biomedical text mining. Bioinf (Oxf). (2020) 36:1234–40. doi: 10.1093/bioinformatics/btz682, 31501885 PatraBG SharmaMM VekariaV AdekkanattuP PattersonOV GlicksbergB . Extracting social determinants of health from electronic health records using natural language processing: a systematic review. J Am Med Inform Assoc. (2021) 28:2716–27. doi: 10.1093/jamia/ocab170, 34613399 CariniC SeyhanAA. Tribulations and future opportunities for artificial intelligence in precision medicine. J Transl Med. (2024) 22:411. doi: 10.1186/s12967-024-05067-0, 38702711 JohnsonKB WeiW-Q WeeraratneD FrisseME MisulisK RheeK . Precision medicine, AI, and the future of personalized health care. Clin Transl Sci. (2021) 14:86–93. doi: 10.1111/cts.12884, 32961010 PerezK WisniewskiD AriA LeeK LieneckC RamamonjiariveloZ. Investigation into application of AI and telemedicine in rural communities: a systematic literature review. Health. (2025) 13:324. doi: 10.3390/healthcare13030324, 39942513 Ciecierski-HolmesT SinghR AxtM BrennerS BarteitS. Artificial intelligence for strengthening healthcare systems in low- and middle-income countries: a systematic scoping review. NPJ Digital Med. (2022) 5:162. doi: 10.1038/s41746-022-00700-y, 36307479 LiuN LiuY LoganB XuZ TangJ WangY. Learning the Dynamic Treatment Regimes from Medical Registry Data through Deep Q-network. Sci Rep. (2019) 9:1495. doi: 10.1038/s41598-018-37142-0, 30728403 WangY LiuA YangJ WangL XiongN ChengY . Clinical knowledge-guided deep reinforcement learning for sepsis antibiotic dosing recommendations. Artif Intell Med. (2024) 150:102811. doi: 10.1016/j.artmed.2024.102811, 38553154 HinrichsN MeyerA KoehlerK KaasT HiddemannM SpethmannS . Artificial intelligence based real-time prediction of imminent heart failure hospitalisation in patients undergoing non-invasive telemedicine. Front Cardiovascular Med. (2024) 11:1457995. doi: 10.3389/fcvm.2024.1457995, 39371396 NovakovskyG DexterN LibbrechtMW WassermanWW MostafaviS. Obtaining genetics insights from deep learning via explainable artificial intelligence. Nat Rev Genet. (2023) 24:125–37. doi: 10.1038/s41576-022-00532-2, 36192604 RakaeeM TafavvoghiM RicciutiB AlessiJV CortelliniA CitarellaF . Deep learning model for predicting immunotherapy response in advanced non-small cell lung cancer. JAMA Oncol. (2025) 11:109–18. doi: 10.1001/jamaoncol.2024.5356, 39724105 DuS XieW GaoS ZhaoR WangH TianJ . Early prediction of neoadjuvant therapy response in breast cancer using MRI-based neural networks: data from the ACRIN 6698 trial and a prospective Chinese cohort. Breast Cancer Res. (2025) 27:52. doi: 10.1186/s13058-025-02009-6, 40181457 HoadleyKA YauC WolfDM CherniackAD TamboreroD NgS . Multiplatform analysis of 12 cancer types reveals molecular classification within and across tissues of origin. Cell. (2014) 158:929–44. doi: 10.1016/j.cell.2014.06.049, 25109877 SomashekharSP SepúlvedaMJ PuglielliS NordenAD ShortliffeEH Rohit KumarC . Watson for oncology and breast cancer treatment recommendations: agreement with an expert multidisciplinary tumor board. Ann Oncol. (2018) 29:418–23. doi: 10.1093/annonc/mdx781, 29324970 CecchinE StoccoG. Pharmacogenomics and Personalized Medicine. Gene. (2020) 11:679. doi: 10.3390/genes11060679, 32580376 SwenJJ NijenhuisM de BoerA GrandiaL Maitland-van der ZeeAH MulderH . Pharmacogenetics: from bench to byte--an update of guidelines. Clin Pharmacol Ther. (2011) 89:662–73. doi: 10.1038/clpt.2011.34, 21412232 GrossiE PoddaGM PuglianoM GabbaS VerriA CarpaniG . Prediction of optimal warfarin maintenance dose using advanced artificial neural networks. Pharmacogenomics. (2014) 15:29–37. doi: 10.2217/pgs.13.212, 24329188 TatonettiNP YePP DaneshjouR AltmanRB. Data-driven prediction of drug effects and interactions. Sci Transl Med. (2012) 4:125ra31. doi: 10.1126/scitranslmed.3003377, 22422992 GenyM AndresE TalhaS GenyB. Liability of health professionals using sensors, telemedicine and artificial intelligence for remote healthcare. Sensors. (2024) 24:491. doi: 10.3390/s24113491, 38894282 KangJH ChangCS NohJJ KimTJ. Does Robot Assisted Laparoscopy (RAL) Have an Advantage in Preservation of Ovarian Reserve in Endometriosis Surgery? Comparison of Single-Port Access (SPA) RAL and SPA Laparoscopy. J Clin Med. (2023) 12:4673. doi: 10.3390/jcm12144673, 37510787 VoigtlaenderS PawelczykJ GeigerM VaiosEJ KarschniaP CudkowiczM . Artificial intelligence in neurology: opportunities, challenges, and policy implications. J Neurol. (2024) 271:2258–73. doi: 10.1007/s00415-024-12220-8, 38367046 DarSU YurtM KaracanL ErdemA ErdemE CukurT. Image Synthesis in Multi-Contrast MRI With Conditional Generative Adversarial Networks. IEEE Trans Med Imaging. (2019) 38:2375–88. doi: 10.1109/tmi.2019.2901750, 30835216 Frid-AdarM DiamantI KlangE AmitaiM GoldbergerJ GreenspanH. GAN-based synthetic medical image augmentation for increased CNN performance in liver lesion classification. Neurocomputing. (2018) 321:321–31. doi: 10.1016/j.neucom.2018.09.013 GoodfellowI Pouget-AbadieJ MirzaM XuB Warde-FarleyD OzairS . Generative adversarial networks. Commun ACM. (2020) 63:139–44. doi: 10.1145/3422622 IslamS AzizMT NabilHR JimJR MridhaMF KabirMM . Generative adversarial networks (GANs) in medical imaging: advancements, applications, and challenges. IEEE Access. (2024) 12:35728–53. doi: 10.1109/ACCESS.2024.3370848 RodriguesT BernardesGJL. Machine learning for target discovery in drug development. Curr Opin Chem Biol. (2020) 56:16–22. doi: 10.1016/j.cbpa.2019.10.003, 31734566 DurhamJ ZhangJ SchaefferRD CongQ. DPAM-AI: a domain parser for AlphaFold models powered by artificial intelligence. Bioinformatics. (2024) 41:740. doi: 10.1093/bioinformatics/btae740, 39672676 AshleyEA. The precision medicine initiative: a new national effort. JAMA. (2015) 313:2119–20. doi: 10.1001/jama.2015.3595, 25928209 ChenY Elenee ArgentinisJD WeberG. IBM Watson: how cognitive computing can be applied to big data challenges in life sciences research. Clin Ther. (2016) 38:688–701. doi: 10.1016/j.clinthera.2015.12.001, 27130797 ManzCR ZhangY ChenK LongQ SmallDS EvansCN . Long-term effect of machine learning-triggered behavioral nudges on serious illness conversations and end-of-life outcomes among patients with cancer: a randomized clinical trial. JAMA Oncol. (2023) 9:414–8. doi: 10.1001/jamaoncol.2022.6303, 36633868 MonlezunDJ SinyavskiyO PetersN SteignerL AksamitT GiraultMI . Artificial intelligence-augmented propensity score, cost effectiveness and computational ethical analysis of cardiac arrest and active cancer with novel mortality predictive score. Medicina (Kaunas). (2022) 58:1039. doi: 10.3390/medicina58081039, 36013506 HampelH VergalloA CaraciF CuelloAC LemercierP VellasB . Future avenues for Alzheimer's disease detection and therapy: liquid biopsy, intracellular signaling modulation, systems pharmacology drug discovery. Neuropharmacology. (2021) 185:108081. doi: 10.1016/j.neuropharm.2020.108081, 32407924 NiaziSK MariamZ. Artificial intelligence in drug development: reshaping the therapeutic landscape. Therapeutic Advan Drug Safety. (2025) 16:704. doi: 10.1177/20420986251321704, 40008227 FischerRP VolpertA AntoninoP AhrensTD. Digital patient twins for personalized therapeutics and pharmaceutical manufacturing. Front Digital Health. (2023) 5:1302338. doi: 10.3389/fdgth.2023.1302338, 38250053 KapadiaK RuwaliS MalavT SeshadriS SeidmannA PressDZ . Enhancing Efficiency with an AI-Augmented Clinician in Neurology. Aging Dis. (2024) 16:2501–3. doi: 10.14336/ad.2024.1249, 39500349 TranBX LatkinCA VuGT NguyenHLT NghiemS TanMX . The current research landscape of the application of artificial intelligence in managing cerebrovascular and heart diseases: a bibliometric and content analysis. Int J Environ Res Public Health. (2019) 16:2699. doi: 10.3390/ijerph16152699, 31362340 ZhangW ZhangJ LiuH LiuY ShengX ZhouS . Functional hydrogel empowering 3D printing titanium alloys. Mater Today Bio. (2025) 30:101422. doi: 10.1016/j.mtbio.2024.101422, 39830135 LiuX FaesL KaleAU WagnerSK FuDJ BruynseelsA . A comparison of deep learning performance against health-care professionals in detecting diseases from medical imaging: a systematic review and meta-analysis. Lancet Digit Health. (2019) 1:e271–97. doi: 10.1016/s2589-7500(19)30123-2 MattisonG CanfellO ForresterD DobbinsC SmithD TöyräsJ . The Influence of Wearables on Health Care Outcomes in Chronic Disease: Systematic Review. J Med Internet Res. (2022) 24:e36690. doi: 10.2196/36690, 35776492 UtshaUT MorshedBI. CardioHelp: a smartphone application for beat-by-beat ECG signal analysis for real-time cardiac disease detection using edge-computing AI classifiers. Smart Health. (2024) 31:100446. doi: 10.1016/j.smhl.2024.100446 ZouQ QuK LuoY YinD JuY TangH. Predicting Diabetes Mellitus With Machine Learning Techniques. Front Genet. (2018) 9:515. doi: 10.3389/fgene.2018.00515, 30459809 SegarMW VaduganathanM PatelKV McGuireDK ButlerJ FonarowGC . Machine learning to predict the risk of incident heart failure hospitalization among patients with diabetes: the WATCH-DM risk score. Diabetes Care. (2019) 42:2298–306. doi: 10.2337/dc19-0587, 31519694 YangX ChenA PourNejatianN ShinHC SmithKE ParisienC . A large language model for electronic health records. NPJ Digit Med. (2022) 5:194. doi: 10.1038/s41746-022-00742-2, 36572766 ZhengNS AnnisJ MasterH HanL GleichaufK ChingJH . Sleep patterns and risk of chronic disease as measured by long-term monitoring with commercial wearable devices in the All of Us Research Program. Nat Med. (2024) 30:2648–56. doi: 10.1038/s41591-024-03155-8, 39030265 LeeHB MeeseepongM TrungTQ KimBY LeeNE. A wearable lab-on-a-patch platform with stretchable nanostructured biosensor for non-invasive immunodetection of biomarker in sweat. Biosens Bioelectron. (2020) 156:112133. doi: 10.1016/j.bios.2020.112133, 32174559 DagherL ShiH ZhaoY MarroucheNF. Wearables in cardiology: Here to stay. Heart Rhythm. (2020) 17:889–95. doi: 10.1016/j.hrthm.2020.02.023, 32354455 SongTA ChowdhurySR MalekzadehM HarrisonS HogeTB RedlineS . AI-driven sleep staging from actigraphy and heart rate. PLoS One. (2023) 18:e0285703. doi: 10.1371/journal.pone.0285703, 37195925 HarroldEC FooteMB RousseauB WalchH KemelY RichardsAL . Neoplasia risk in patients with Lynch syndrome treated with immune checkpoint blockade. Nat Med. (2023) 29:2458–63. doi: 10.1038/s41591-023-02544-9, 37845474 TruongATL TanLWJ ChewKA VillarazaS SiongcoP BlasiakA . Harnessing CURATE.AI for N-of-1 optimization analysis of combination therapy in hypertension patients: a retrospective case series. Adv Ther. (2021) 4:2100091. doi: 10.1002/adtp.202100091 TianY LiuX WangZ CaoS LiuZ JiQ . Concordance between watson for oncology and a multidisciplinary clinical decision-making team for gastric cancer and the prognostic implications: retrospective study. J Med Internet Res. (2020) 22:e14122. doi: 10.2196/14122, 32130123 ShajariS KuruvinashettiK KomeiliA SundararajU. The emergence of AI-based wearable sensors for digital health technology: a review. Sensors (Basel, Switzerland). (2023) 23:498. doi: 10.3390/s23239498, 38067871 SenadheeraI HettiarachchiP HaslamB NawaratneR SheehanJ LockwoodKJ . AI applications in adult stroke recovery and rehabilitation: a scoping review using AI. Sensors (Basel, Switzerland). (2024) 24:585. doi: 10.3390/s24206585, 39460066 LiD VaidyaJ WangM BushB LuC KollefM . Feasibility study of monitoring deterioration of outpatients using multimodal data collected by wearables. ACM Transactions on Computing for Healthcare. (2020) 1:3344256. doi: 10.1145/3344256 GhorbaniF AhmadiA KiaM RahmanQ DelrobaeiM. A decision-aware ambient assisted living system with IoT embedded device for in-home monitoring of older adults. Sensors (Basel, Switzerland). (2023) 23:673. doi: 10.3390/s23052673, 36904877 TannerJ RochonM HarrisR BeckhellingJ JurkiewiczJ MasonL . Digital wound monitoring with artificial intelligence to prioritise surgical wounds in cardiac surgery patients for priority or standard review: protocol for a randomised feasibility trial (WISDOM). BMJ Open. (2024) 14:e086486. doi: 10.1136/bmjopen-2024-086486, 39289023 HaleemA JavaidM SinghRP SumanR. Telemedicine for healthcare: Capabilities, features, barriers, and applications. Sens Int. (2021) 2:100117. doi: 10.1016/j.sintl.2021.100117, 34806053 KwongJCC KhondkerA LajkoszK McDermottMBA FrigolaXB McCraddenMD . APPRAISE-AI tool for quantitative evaluation of AI studies for clinical decision support. JAMA Netw Open. (2023) 6:e2335377. doi: 10.1001/jamanetworkopen.2023.35377, 37747733 WangX WeiM ChenY JiaJ ZhangY DaiY . Intratumoral and peritumoral MRI-based radiomics for predicting extrapelvic peritoneal metastasis in epithelial ovarian cancer. Insights Imag. (2024) 15:281. doi: 10.1186/s13244-024-01855-w, 39576435 SoltanAAS YangJ PattanshettyR NovakA YangY RohanianO . Real-world evaluation of rapid and laboratory-free COVID-19 triage for emergency care: external validation and pilot deployment of artificial intelligence driven screening. Lancet Digital Health. (2022) 4:e266–78. doi: 10.1016/S2589-7500(21)00272-7, 35279399 ChamelyEA KaoLS. Individualising the recovery process through eHealth. Lancet Digital Health. (2023) 5:e478–9. doi: 10.1016/S2589-7500(23)00098-5, 37419844 BergerMF PetritschJ HeckerA PustakS MichelitschB BanfiC . Paper-and-pencil vs. electronic patient records: analyzing time efficiency, personnel requirements, and usability impacts on healthcare administration. J Clin Med. (2024) 13:6214. doi: 10.3390/jcm13206214, 39458164 ChenJ LiuY LiuP ZhaoY ZuoY DuanH. Adoption of Large Language Model AI Tools in Everyday Tasks: Multisite Cross-Sectional Qualitative Study of Chinese Hospital Administrators. J Med Internet Res. (2025) 27:e70789. doi: 10.2196/70789, 40116330 Tanguay-SelaM BenrimohD PopescuC PerezT RollinsC SnookE . Evaluating the perceived utility of an artificial intelligence-powered clinical decision support system for depression treatment using a simulation center. Psychiatry Res. (2022) 308:114336. doi: 10.1016/j.psychres.2021.114336, 34953204 AggarwalR SounderajahV MartinG TingDSW KarthikesalingamA KingD . Diagnostic accuracy of deep learning in medical imaging: a systematic review and meta-analysis. NPJ Digit Med. (2021) 4:65. doi: 10.1038/s41746-021-00438-z, 33828217 AttiaZI NoseworthyPA Lopez-JimenezF AsirvathamSJ DeshmukhAJ GershBJ . An artificial intelligence-enabled ECG algorithm for the identification of patients with atrial fibrillation during sinus rhythm: a retrospective analysis of outcome prediction. Lancet. (2019) 394:861–7. doi: 10.1016/s0140-6736(19)31721-0, 31378392 McKinneySM SieniekM GodboleV GodwinJ AntropovaN AshrafianH . International evaluation of an AI system for breast cancer screening. Nature. (2020) 577:89–94. doi: 10.1038/s41586-019-1799-6 KellyCJ KarthikesalingamA SuleymanM CorradoG KingD. Key challenges for delivering clinical impact with artificial intelligence. BMC Med. (2019) 17:195. doi: 10.1186/s12916-019-1426-2, 31665002 FinlaysonSG SubbaswamyA SinghK BowersJ KupkeA ZittrainJ . The clinician and dataset shift in artificial intelligence. N Engl J Med. (2021) 385:283–6. doi: 10.1056/NEJMc2104626, 34260843 EdemekongPF AnnamarajuP AfzalM HaydelMJ. Health Insurance Portability and Accountability Act (HIPAA) Compliance. Florida: StatPearls Publishing (2024). HoofnagleCJ van der SlootB BorgesiusFZ. The European Union general data protection regulation: what it is and what it means*. Information Commun Technol Law. (2019) 28:65–98. doi: 10.1080/13600834.2019.1573501 PatiS KumarS VarmaA EdwardsB LuC QuL . Privacy preservation for federated learning in health care. Patterns. (2024) 5:100974. doi: 10.1016/j.patter.2024.100974, 39081567 AouediO SaccoA PiamratK MarchettoG. Handling privacy-sensitive medical data with federated learning: challenges and future directions. IEEE J Biomed Health Inform. (2023) 27:790–803. doi: 10.1109/JBHI.2022.3185673, 35737624 ChoiSJ ChenM TanX. Assessing the impact of health information exchange on hospital data breach risk. Int J Med Inform. (2023) 177:105149. doi: 10.1016/j.ijmedinf.2023.105149, 37453177 GhassemiM Oakden-RaynerL BeamAL. The false hope of current approaches to explainable artificial intelligence in health care. Lancet Digital Health. (2021) 3:e745–50. doi: 10.1016/S2589-7500(21)00208-9, 34711379 HarrerS. Attention is not all you need: the complicated case of ethically using large language models in healthcare and medicine. EBioMedicine. (2023) 90:104512. doi: 10.1016/j.ebiom.2023.104512, 36924620 ReddyS. Evaluating large language models for use in healthcare: a framework for translational value assessment. Inf Med Unlocked. (2023) 41:101304. doi: 10.1016/j.imu.2023.101304 AmannJ BlasimmeA VayenaE FreyD MadaiVI. Explainability for artificial intelligence in healthcare: a multidisciplinary perspective. BMC Med Inform Decis Mak. (2020) 20:310. doi: 10.1186/s12911-020-01332-6, 33256715 KarakoK TangW. Applications of and issues with machine learning in medicine: Bridging the gap with explainable AI. Biosci Trends. (2025) 18:497–504. doi: 10.5582/bst.2024.01342, 39647859 KiyassehD LacaJ HaqueTF OtiatoM MilesBJ WagnerC . Human visual explanations mitigate bias in AI-based assessment of surgeon skills. NPJ Digit Med. (2023) 6:54. doi: 10.1038/s41746-023-00766-2, 36997642 MittermaierM RazaMM KvedarJC. Bias in AI-based models for medical applications: challenges and mitigation strategies. NPJ Digital Med. (2023) 6:113. doi: 10.1038/s41746-023-00858-z, 37311802 ChinMH Afsar-ManeshN BiermanAS ChangC Colon-RodriguezCJ DullabhP . Guiding principles to address the impact of algorithm bias on racial and ethnic disparities in health and health care. JAMA Netw Open. (2023) 6:e2345050. doi: 10.1001/jamanetworkopen.2023.45050, 38100101 LandersRN BehrendTS. Auditing the AI auditors: A framework for evaluating fairness and bias in high stakes AI predictive models. Am Psychol. (2023) 78:36–49. doi: 10.1037/amp0000972, 35157476 LiuM NingY TeixayavongS MertensM XuJ TingDSW . A translational perspective towards clinical AI fairness. NPJ Digit Med. (2023) 6:172. doi: 10.1038/s41746-023-00918-4, 37709945 ParvinenL AlamäkiA HallikainenH MäkiM. Exploring the challenges of and solutions to sharing personal genomic data for use in healthcare. Health Inform J. (2023) 29:14604582231152185. doi: 10.1177/14604582231152185, 36651319 KokJ. W. T. M.de HozM. Á. A. de JongY. BrokkeV. 2023 A guide to sharing open healthcare data under the General Data Protection Regulation Sci Data 10:404 doi: 10.1038/s41597-023-02256-2, 37355751 ČartolovniA TomičićA Lazić MoslerE. Ethical, legal, and social considerations of AI-based medical decision-support tools: A scoping review. Int J Med Inform. (2022) 161:104738. doi: 10.1016/j.ijmedinf.2022.104738, 35299098 Mohsin KhanM ShahN ShaikhN ThabetA AlrabayahT BelkhairS. Towards secure and trusted AI in healthcare: a systematic review of emerging innovations and ethical challenges. Int J Med Inform. (2025) 195:105780. doi: 10.1016/j.ijmedinf.2024.105780 Economou-ZavlanosNJ BessiasS CaryMP BedoyaAD GoldsteinBA JelovsekJE . Translating ethical and quality principles for the effective, safe and fair development, deployment and use of artificial intelligence technologies in healthcare. J Am Med Inform Assoc. (2024) 31:705–13. doi: 10.1093/jamia/ocad221, 38031481 ReddyS. Generative AI in healthcare: an implementation science informed translational path on application, integration and governance. Implementation Sci: IS. (2024) 19:27. doi: 10.1186/s13012-024-01357-9, 38491544 SilcoxC ZimlichmannE HuberK RowenN SaundersR McClellanM . The potential for artificial intelligence to transform healthcare: perspectives from international health leaders. NPJ Digit Med. (2024) 7:88. doi: 10.1038/s41746-024-01097-6, 38594477 YangY LinM ZhaoH PengY HuangF LuZ. A survey of recent methods for addressing AI fairness and bias in biomedicine. J Biomed Inform. (2024) 154:104646. doi: 10.1016/j.jbi.2024.104646, 38677633 DongM ConwayJR BonnefonJF ShariffA RahwanI. Fears about artificial intelligence across 20 countries and six domains of application. Am Psychol. (2024). doi: 10.1037/amp0001454, 39666499 MalerbiFK NakayamaLF Gayle DychiaoR Zago RibeiroL VillanuevaC CeliLA . Digital education for the deployment of artificial intelligence in health care. J Med Internet Res. (2023) 25:e43333. doi: 10.2196/43333, 37347537 LambertSI MadiM SopkaS LenesA StangeH BuszelloC-P . An integrative review on the acceptance of artificial intelligence among healthcare professionals in hospitals. NPJ Digit Med. (2023) 6:111. doi: 10.1038/s41746-023-00852-5, 37301946 FisherS RosellaLC. Priorities for successful use of artificial intelligence by public health organizations: a literature review. BMC Public Health. (2022) 22:2146. doi: 10.1186/s12889-022-14422-z, 36419010 AgarwalR BjarnadottirM RhueL DugasM CrowleyK ClarkJ . Addressing algorithmic bias and the perpetuation of health inequities: an AI bias aware framework. Health Policy Technol. (2023) 12:100702. doi: 10.1016/j.hlpt.2022.100702 RatwaniRM SuttonK GalarragaJE. Addressing AI Algorithmic Bias in Health Care. JAMA. (2024) 332:1051–2. doi: 10.1001/jama.2024.13486, 39230911 TopolEJ. High-performance medicine: the convergence of human and artificial intelligence. Nat Med. (2019) 25:44–56. doi: 10.1038/s41591-018-0300-7, 30617339 CabitzaF RasoiniR GensiniGF. Unintended Consequences of Machine Learning in Medicine. JAMA. (2017) 318:517–8. doi: 10.1001/jama.2017.7797, 28727867 MennellaC ManiscalcoU De PietroG EspositoM. Ethical and regulatory challenges of AI technologies in healthcare: A narrative review. Heliyon. (2024) 10:e26297. doi: 10.1016/j.heliyon.2024.e26297, 38384518 PalaniappanK LinEYT VogelS LimJCW. Gaps in the global regulatory frameworks for the use of artificial intelligence (AI) in the healthcare services sector and key recommendations. Health. (2024) 12:730. doi: 10.3390/healthcare12171730, 39273754 Sousa-PintoB Marques-CruzM NeumannI ChiY NowakAJ ReinapM . Guidelines International Network: principles for use of artificial intelligence in the health guideline enterprise. Ann Intern Med. (2025) 178:408–15. doi: 10.7326/ANNALS-24-02338 FedericoCA TrotsyukAA. Biomedical Data Science, Artificial Intelligence, and Ethics: Navigating Challenges in the Face of Explosive Growth. Annu Rev Biomed Data Sci. (2024) 7:1–14. doi: 10.1146/annurev-biodatasci-102623-104553, 38598860 WilliamsM KarimW GelmanJ RazaM. Ethical data acquisition for LLMs and AI algorithms in healthcare. NPJ Digit Med. (2024) 7:377. doi: 10.1038/s41746-024-01399-9, 39715803 AbbasS QaisarA FarooqMS SaleemM AhmadM KhanMA. Smart vision transparency: efficient ocular disease prediction model using explainable artificial intelligence. Sensors. (2024) 24:6618. doi: 10.3390/s24206618, 39460097 HassanM KushnirukA BoryckiE. Barriers to and Facilitators of Artificial Intelligence Adoption in Health Care: Scoping Review. JMIR Hum Factors. (2024) 11:e48633. doi: 10.2196/48633, 39207831 KauttonenJ RousiR AlamäkiA. Trust and Acceptance Challenges in the Adoption of AI Applications in Health Care: Quantitative Survey Analysis. J Med Internet Res. (2025) 27:e65567. doi: 10.2196/65567, 40116853 SaenzAD CentiA TingD YouJG LandmanA MishurisRG. Establishing responsible use of AI guidelines: a comprehensive case study for healthcare institutions. NPJ Digit Med. (2024) 7:348. doi: 10.1038/s41746-024-01300-8, 39616269 SadilekA LiuL NguyenD KamruzzamanM SerghiouS RaderB . Privacy-first health research with federated learning. NPJ Digit Med. (2021) 4:132. doi: 10.1038/s41746-021-00489-2, 34493770 US. Food and Drug Administration (2021) Artificial Intelligence/Machine Learning (AI/ML)-Based Software as a Medical Device (SaMD) Action Plan U.S. Food and Drug Administration Available online at: https://www.fda.gov/media/145022/download (Accessed November 18, 2025). European Parliament, & Council of the European Union. Regulation (EU) 2024/1689 on Artificial Intelligence (AI Act). Official J European Union, L 2024/1689. (2024) 1–144. Available at: http://data.europa.eu/eli/reg/2024/1689/oj World Health Organization. Ethics and Governance of Artificial Intelligence for Health: WHO Guidance. Geneva: World Health Organization (2021). RajiID BuolamwiniJ. Actionable auditing revisited-investigating the impact of publicly naming biased performance results of commercial AI products [Article]. Commun ACM. (2023) 66:101–8. doi: 10.1145/3571151 VaseyB NagendranM CampbellB CliftonDA CollinsGS DenaxasS . Reporting guideline for the early stage clinical evaluation of decision support systems driven by artificial intelligence: DECIDE-AI. BMJ. (2022) 377:e070904. doi: 10.1136/bmj-2022-070904, 35584845 MorleyJ MachadoCCV BurrC CowlsJ JoshiI TaddeoM . The ethics of AI in health care: A mapping review. Soc Sci Med. (2020) 260:113172. doi: 10.1016/j.socscimed.2020.113172, 32702587 DesoldaG EspositoA LanzilottiR PiccinnoA CostabileMF. From human-centered to symbiotic artificial intelligence: a focus on medical applications. Multimed Tools Appl. (2024) 84:414. doi: 10.1007/s11042-024-20414-5 LemboS. BarraP. DashS. R. BiasiL. D. (2024). “Challenges and Opportunities of Symbiotic AI in Rare Disease Diagnosis.” 2024 IEEE International Conference on Bioinformatics and Biomedicine (BIBM). GlickmanM SharotT. AI-induced hyper-learning in humans. Curr Opin Psychol. (2024) 60:101900. doi: 10.1016/j.copsyc.2024.101900, 39348730 MensahA BaoQ ZhangZ ChenY JiangQ CaiP. Symbiopersonal intelligence towards symbiotic and personalized digital medicine. Fundam Res. (2025) 5:1423–8. doi: 10.1016/j.fmre.2025.01.009, 40777803 YuanL GaoX ZhengZ EdmondsM WuYN RossanoF . In situ bidirectional human-robot value alignment. Sci Robot. (2022) 7:eabm4183. doi: 10.1126/scirobotics.abm4183, 35857532 LoftusTJ ShickelB Ozrazgat-BaslantiT RenY GlicksbergBS CaoJ . Artificial intelligence-enabled decision support in nephrology. Nat Rev Nephrol. (2022) 18:452–65. doi: 10.1038/s41581-022-00562-3, 35459850 TschandlP RinnerC ApallaZ ArgenzianoG CodellaN HalpernA . Human–computer collaboration for skin cancer recognition. Nat Med. (2020) 26:1229–34. doi: 10.1038/s41591-020-0942-0 MolluraDJ CulpMP PollackE BattinoG ScheelJR MangoVL . Artificial intelligence in low- and middle-income countries: innovating global health radiology. Radiology. (2020) 297:513–20. doi: 10.1148/radiol.2020201434, 33021895 PirmaniA OldenhofM PeetersLM De BrouwerE MoreauY. Accessible Ecosystem for Clinical Research (Federated Learning for Everyone): Development and Usability Study. JMIR Formative Res. (2024) 8:e55496. doi: 10.2196/55496, 39018557 XuW ChenB ShiH TianH XuX. Real-time COVID-19 detection over chest x-ray images in edge computing. Comput Intell. (2022). doi: 10.1111/coin.12528, 35941908

Edited by: Guglielmo M. Trovato, European Medical Association (EMA), Belgium

Reviewed by: Luigi Di Biasi, University of Salerno, Italy

Suhas Srinivasan, Stanford University, United States

Dinesh Mendhe, Rutgers Robert Wood Johnson University Hospital, United States