Evolving surgical teams in the age of artificial intelligence and robotics

Abstract

Surgery is a critical function of the healthcare system, key to addressing a substantial portion of the global disease burden. The integration of advanced artificial intelligence (AI) and robotics ecosystems into the operating room (OR) promises to radically transform surgery, with profound implications. This article analyzes the current state of surgical AI and robotic systems; presents a vision for their future, highlighting technological and research challenges and their associated impact on surgical teams; and discusses the ethical and regulatory implications. AI systems will use complex, multimodal data streams collected from patients, surgical teams, robots, and the OR environment to become increasingly capable of situational awareness, workflow recognition, performance benchmarking, causal inference, outcome prediction, and intraoperative decision-making to optimize surgical actions. Robotics will move from passive instrument-handling tools to autonomous systems with human-in-the-loop control, with embodied AI and enhanced sensor-based perception providing comprehensive spatial–temporal understanding, anticipatory behaviors, and adaptive learning. The surgeon’s role will shift toward supervision, coordination, and high-level decision-making, while nurses, assistants, and anesthesiologists will have additional competencies complemented by clinical data scientists and AI and robotic integration engineers. Ethical challenges will include liability and the implications of diluted authority chains, the potential for AI bias to exacerbate health inequalities, and the concentration of research and industry in resource-rich nations. New regulatory and compliance frameworks, trial methods, reporting standards, and training approaches will be needed to ensure the safety and effectiveness of these systems. Ultimately, AI and robotics should sustain, rather than disrupt, surgical practices by refining the skills of care providers to achieve true personalized surgery and propel procedural and technological advancements.

Key points

Introduction

Surgery is a critical component of the healthcare system, addressing a substantial portion of the global disease burden. This portion is difficult to quantify but is estimated to be approximately 30% by those providing healthcare (1). Overall, 28.6% of hospital inpatients undergo major surgery, with the highest rates among those admitted with musculoskeletal (84.0%) and neoplastic (61.4%) primary diagnoses (2). Despite past improvements in the quality of surgical care due to innovations in diagnostic and therapeutic technologies, surgery faces global challenges (3). These challenges include a gap of approximately 143.1 million unmet surgical procedures per year (4), an expected 52% increase in neoplastic cancerous conditions requiring surgical treatment between 2018 and 2040 (5), surgical complications being the third most common cause of death worldwide (6), access to safe and effective surgery (3), and healthcare inequities, particularly related to cancer surgery (7).

Surgical robots have been introduced over the last two decades with a strong focus on improving the technical performance of procedures (8). More recently, the evolution of artificial intelligence (AI) has raised questions about the joint role of AI and robotics in the operating room (OR) ecosystem and throughout patients’ journeys—both now and in the coming years. Although current technological trends highlight the importance of “human-in-the-loop” approaches, the advent of autonomous robotics and AI (9, 10) means that it is unclear how one can maximize the effectiveness of human–robot collaboration in the OR. This issue goes beyond the role of surgeons and has ethical and practical implications for the entire surgical team.

The application of AI to suitable datasets has the potential to transform surgical care. Progress to date in surgical data science (11), primarily related to the analysis of surgical videos, offers only a glimpse of this future. Recent developments in causality, large language models (LLMs), embodied AI, world models, and autonomy promise to revolutionize robotic surgery entirely. However, this revolution will not meaningfully take place until we address key issues related to data scarcity, privacy, ethics, data equity, and bias—all factors that affect the generalizability, robustness, and trustworthiness of models.

Given the enormous societal implications, there is a pressing need for discussion encompassing all relevant stakeholders to explore the potential opportunities and challenges that AI and robotics present in surgery. This article aims to inform this process by describing the underlying progress in the technologies that support AI and robotics, which will allow surgery to better address the global disease burden and the associated implications for human factors. First, we outline how the role of the surgical team has evolved to date in the context of innovation. Next, we analyze the current state of the art of AI and robotic systems in the OR and present a vision for their future, highlighting technological and research challenges and the anticipated impact on surgical teams. Finally, we discuss the associated implications for ethics, equity, and regulation.

Surgical team evolution through innovation

In order to anticipate/predict the likely impact of AI and robotics on the future of surgery and surgical teams, it is important to understand how innovation has shaped today’s surgical teams, interventions, and outcomes.

Surgical innovation

Surgery has evolved through continuous innovation within a unique culture and deeply rooted tradition in a landscape shaped by significant health policy considerations (12). Historically, surgeons have been leading innovators, anticipating advances and seizing opportunities through their deep understanding of clinical needs—a role that remains vital as surgery undergoes rapid and accelerating transformation (13). Innovations that have revolutionized surgery include instrument sterilization, which has supported countless advances; anesthesia, which has enabled surgeons to operate without patients experiencing severe pain; antiseptic surgery, which has reduced the longstanding threat of infection; and vascular anastomosis, which has promoted a series of advances in vascular repair and organ transplantation, among others (12, 14). Over the past few decades, minimally invasive surgery has been the key driver of innovation. However, compared to these enabling technologies, AI and robotics are likely to be disruptive, transforming the way we see and experience surgery and reshaping the roles of those providing care in the OR.

Surgical teams

Surgical teams include surgeons, anesthesiologists, nurses, perfusionists, surgical trainees, anesthetic practitioners, and physician assistants (15). Traditionally, their interactions are subject to their relative physical positioning and are characterized by formal and informal communication, both verbal and nonverbal (13). Surgical teams may adapt to a particular surgeon’s personal style through a common understanding of expectations and unspoken cues; for instance, each scrub nurse anticipates individual surgeons differently. They do so by sharing common mental models, with this implicit culture leading to increased efficiency in surgical practice. However, discrepant views of which behaviors enhance or undermine teamwork in the OR are pervasive (15), a problem exacerbated by the observer’s own preferences and past experience (16). Moreover, although leadership cultivates teamwork, inclusivity, and engagement in the OR (17), it has traditionally been surgeon-centered. Leadership reflects a balance of roles, such as delegator, elucidator, and tone setter—each varying in influence and perceived value (18). Surgeons are also the primary decision-makers in the OR, mitigating risks on a continuous basis (12). However, at times, a surgeon’s mentality as the sole decision-maker may lead to catastrophic events by failing to address other team members’ concerns.

Although teamwork, leadership, and decision-making reflect how surgical teams work, their importance remains largely underappreciated (19, 20). As large language and reasoning models mature, technology could fill those gaps by playing a supportive role, for example, by prompting the members of a team, carrying out accurate simulations, and implementing checklists behind the scenes (21, 22). AI could indirectly improve leadership by augmenting surgeons’ situational awareness and reducing their cognitive load, supporting clearer and faster communication of vital information to other members of the team. For instance, AI could inform the surgical team about the current surgical workflow (23), allowing them to anticipate tasks jointly (24) and reducing variation in care—as seen in the National Health Service (NHS) England’s Getting It Right First Time (GIRFT) program (25)—thereby improving the efficiency and quality of the procedure. However, any reduction in human-to-human engagement may have an impact on trust within teams. AI will also allow for increased equity in information among other members of the surgical team, sharing more responsibility for decisions made in the OR. However, when these decisions are the product of an AI model, this could lead to confusion about who is “in charge,” as surgeon authority may be diluted.

Surgical interventions

Regardless of how minimal interventions have become, a surgical procedure is fundamentally represented by an image and a manipulation, carefully coordinated by a surgical team (26). Until the 1980s, surgery was surgeon-centric (13): other team members were unable to see or do as much as the surgeon owing to limitations in space. Innovations that overcame this limitation included camera-based, minimally invasive procedures that enabled the surgical team to see exactly what the surgeon was seeing and allowed other team members to assume more responsibilities. For example, nurses could be upskilled and assistants could be given more active roles. A similar example can be seen in microsurgery (e.g., neurosurgery or ophthalmology): in these procedures, the surgeon’s perception, coordination, and interactions are routed through an intermediary optical or navigational interface between the surgeon and the patient, reshaping intraoperative communication and team dynamics. These two examples contrast with open surgery, in which typically only the surgeon and their assistant(s) have a direct view of the operative field. Moreover, a very apparent change in robotic-assisted surgery is the positioning of the surgeon at a console rather than at the patient’s bedside. This impacts interpersonal relationships with other team members, especially in closed-console designs (e.g., da Vinci robotic systems; Intuitive Surgical, Inc.). Finally, computational tools increasingly support procedural planning in stereotactic neurosurgery and orthopedics, opening information to the rest of the surgical team even before starting an intervention. Across specialties, the key disruption is, therefore, not only about automation and decision-making support but also about a redistribution of perception and planning. This changes who possesses information, who anticipates the next step, and how the team coordinates.

What surgical team members see and do will inevitably be challenged by AI and robotics. The problem is that doing has become unfashionable, even in surgery (27). “We have become mind rich and body poor” (28) because we believe that mental activities require more intelligence than practical physical ones. However, if AI is able to capture and analyze large amounts of knowledge and pave the way for automation, then what functions will be left for the surgical team?

Surgical outcomes

Since the 1980s, surgical teams have gradually evolved from being surgeon-dominant to a multidisciplinary structure in which other members of the team are empowered to advocate for patient safety. There was recognition that thousands of deaths globally were caused by preventable medical errors and that systemic change was needed (29). Beyond technical competency, it became evident that surgical team dynamics can also have a profound impact on patient outcomes. For instance, teams with poor, unstructured communication and coordination have higher rates of postoperative complications and mortality (20). Professional bodies began mandating standardized checklists (e.g., the preoperative time out), briefings, and handovers. Patient advocacy has increased through individual empowerment, access to information, and organized patient safety movements that began examining team failures more closely. A broader shift in societal attitudes began to challenge the perceived infallibility and authority of traditional, paternalistic surgeons. There was also a drive to “upskill” team members, such as nurses, to take on more independent, specialized roles in perioperative care. Surgeons also began depending on intraoperative input from other doctors, such as pathologists and radiologists, emphasizing that collective expertise improves outcomes. Despite this evolution, several aspects of team dynamics have remained the same. The surgeon (or, depending on the situation, the anesthesiologist) is still ultimately responsible for patient safety and the eventual outcome.

Human factors

Today, surgical team members, particularly those with higher responsibilities, may have a negative view of the continuous AI-powered recording and evaluation of their actions and outcomes. This “shadowing” of their practice carries potential risks, such replacing the human element with technology, discouraging surgical teams from taking on high-risk cases, and reinforcing the belief that the ultimate goal is to replace team members with fully autonomous systems (Table 1). In the years ahead, it is unclear what new roles are likely to develop in the OR, beyond surgical robotic company representatives monitoring their systems under evaluation. The introduction of new roles to the traditional surgical team will pose challenges as new disciplines, including data science, become increasingly integrated into the clinical environment.

Surgical artificial intelligence

The application of AI to surgery is an active area of research spanning many fields, including computer vision, surgical data science (11), computer-assisted interventions (30), medical imaging, autonomous robotics (31), and statistics (Table 2). For instance, today, Touch Surgery (Medtronic Ltd.) is able to automatically annotate a surgical video and email it to the surgeon 5 min after completing the case. However, the impact of AI in surgery is limited thus far, as models are trained on too few cases (in the order of dozens), impairing generalizability. This problem results from a lack of representative, heterogeneous, diverse, multimodal, high-dimensional, and annotated data. Currently, curating clinical datasets is a time-consuming effort commonly done by trainees. Data are typically modeled in isolation at specific points of care rather than across entire pathways, i.e., from diagnosis to outcomes. Current datasets primarily comprise data collected before (i.e., demographics, medical images, and diagnoses) or after (i.e., outcomes, complications, morbidity, mortality, and long-term follow-ups) surgery; data collected during surgery are scarce and primarily limited to endoscopic anatomical views (32). Although operating notes are common practice for documenting important events during surgery, this process is subjective, not comprehensive, and underreported. Other modalities, such as audio, robot kinematics and dynamics, and physiological signals (e.g., electrodermal activity), have been investigated for event detection (33), skills assessment (34), and cognitive workload quantification (35), among other applications.

One of the fundamental contributions of current research in surgical AI is related to workflow and scene understanding from surgical videos (36). The goal of these models is to learn optimal feature representations at the scene level (space) during surgery (time). Feature representation learning has been proposed at different levels of granularity, ranging from recognizing surgical phase/step transitions (23, 37) to recognizing surgical gestures (38) and actions (39), in addition to identifying go/no-go zones (40) as a form of intraoperative guidance. Recent approaches have unified these different levels of abstraction into single architectures (41). This video-centric focus has been extended to the analysis of surgical team performance with cameras placed in the OR for three-dimensional human pose estimation and semantic segmentation (44).

Vision

In the following sections, we will outline our vision for the future application of AI in surgery via data science, workflow understanding, “causal” surgery, and surgical co-pilots, together with their implications for the surgical team (Figure 1).

Figure 1

Surgical data science

AI models applied to surgery will move away from processing only endoscopic views of the patient; they will holistically integrate, analyze, and learn from data from different modalities and run inference locally, in the cloud, or in a distributed manner using a combination of high-performance and edge-computing devices (Figure 2). This implies that the OR has the potential to become a highly sensorized, situationally aware space that provides functions such as those of a flight data recorder augmented with near-miss warning systems. However, challenges to be addressed include data synchronization, storage, diversity, privacy, real-time inference, and unclear evidence of the impact of each modality. Efforts underway to solve these and other issues include frameworks—either private (e.g., Proximie; Proximie; Black Box, Surgical Safety Technologies Inc.; and Rods&Cones, Rods&Cones) or open—for automated multimodal data recording and inference in the OR (42) and federated learning (82) with differential privacy (84) for the efficient orchestration and aggregation of data without the need for sharing data outside organizations. Data should be agnostic to medical device providers (76), while guaranteeing interoperability via open standards for data communication and retrieval (85). AI approaches that efficiently document surgery in a qualitative and quantitative manner with automated, large-scale annotations will add value to the surgical team (86).

Figure 2

Surgical data science will be key to linking preoperative and postoperative data, as the current methodology works in silos of data pools that do not span the journeys experienced by patients undergoing surgery. Achieving this goal with current practices is time-consuming and not scalable. More efficient tools are needed for the automated curation and linkage of patient-specific data across healthcare pathways, across diverse clinical studies, and over time, i.e., longitudinally. This continuous and iterative data curation mechanism will become a key enabler for providing evidence from previous experience in the OR, while addressing challenges associated with randomized controlled trials in surgery, since they are long and expensive to execute.

Understanding surgical workflows

Surgery comprises workflows that are extremely complex and highly heterogeneous (30). Surgery will rely on foundation models trained across multi-institutional and diverse surgical datasets and fine-tuned for specific downstream tasks in a sustainable manner. Eventually, these models will consider other modalities, including audio, robot kinematics, and other forms of physiological sensors. The models will primarily be extended to content-coded human speech and communication processing, allowing for the quantification of interactions among members of the surgical team (49). The ultimate goal of these systems will be to ensure high global standards of surgical practice by providing system-agnostic benchmarking of both technical and non-technical surgical performance. This includes aspects such as instrument dexterity, respect for tissue, communication skills, decision-making, and coping with stress. Additionally, these systems will aim to offer real-time intraoperative guidance to manage the rare occurrence of adverse events, both those observed in the scene in real time and those that can be anticipated before they occur (24). Moreover, automatic gesture analysis and performance metrics derived from AI models will not only benchmark competency but also provide important insights regarding postoperative recovery, as has already been demonstrated with clinical outcomes and other patient-centered metrics (38, 87). AI models will require mechanisms to extract the exact three-dimensional location of lesions and contextual surrounding anatomy, along with aggressiveness scores resulting from tool–tissue interaction, allowing for a comprehensive analysis of performance rather than an analysis of gestures in isolation. In addition to performance, automated and systematic approaches are required for the identification of adverse events within better evaluation frameworks to document rare events occurring in the OR (88).

Causal surgery

Along with correlations learned with AI models, the introduction of causal discovery, causal representation learning, causal interventions, and counterfactual learning will affect the data available in the OR, jointly with data from before and after surgery. These causal mechanisms will be able to model causes and effects efficiently, while accounting for spurious correlations and observable and unobservable confounders. Causal discovery typically relies on directed graphs and allows causal information to be inferred from purely observational data instead of relying on expensive and time-consuming randomized experiments (89). Finding optimal representation learning in a causal manner will allow us to better map the underlying data generation process. Causal interventions will allow us to investigate the effect of changes to features on outcomes from a data perspective, rather than enacting actual interventions. Counterfactuals address “what if” questions through retrospective thinking, i.e., by using algorithms designed according to the observed world to produce an answer about a counterfactual world. The convergence of causal inference and neural networks—a field known as causal AI—is a more recent endeavor (90, 91). However, new causal machine learning methods are needed to address real-world considerations, such as the scalability, interpretability, and generalizability of high-dimensional data. Beyond predicting the outcome of a surgical intervention, causal AI will enable the surgical approach to be optimized for patient-specific cases, i.e., prescriptive interventions.

Surgical co-pilots

LLMs are compelling generative models that empower human–computer interaction, consolidate foundation models and downstream tasks, integrate different sources of information, provide solutions for unstructured data via prompt engineering, and enable reasoning capabilities. LLMs are redefining healthcare and accelerating medical research (92), although with challenges that include data privacy, algorithmic bias, fairness, and resistance to adoption (93). In the context of surgery, LLMs provide a framework for co-production, data retrieval, and support for decision-making in the form of a surgical co-pilot. However, LLMs pose risks and challenges that need to be addressed before they can be deployed safely and efficiently in the OR. These challenges and risks include (among others) the dominance of private companies; a lack of transparency; a propensity to provide false information (confabulations, most popularly known as “hallucinations”); the involvement of multiple parties in an “AI chain” (from development to provision and deployment); and the undermining of the epistemic authority of humans, among many others (94). LLMs based on retrieval-augmented generation have been used in the context of surgical data to overcome issues of hallucinations and out-of-context information (95).

Recent extensions of LLMs include vision-language models (VLMs) and vision-language-action (VLA) models. VLMs expand the capabilities of conventional text-based LLMs with vision-based multimodal understanding of surgical environments (46, 96) while supporting natural language explanations of their reasoning processes (97, 98), addressing transparency—one of the key challenges in the deployment of LLMs in clinical settings. VLA models further extend VLMs with an action-generation component by incorporating robotic control capabilities based on visual and textual input, showcasing the potential for autonomous surgical assistance, where natural language instructions from surgical teams can be directly translated into coordinated robotic movements (99, 100).

Surgical decision-making will be supported by AI-based reasoning models that will explore all possible scenarios on behalf of the surgical team. AI models with reasoning capabilities will be able to learn chains of thought based on observations made in real time in the OR while abstracting representations from previous cases. These capabilities will not only provide the surgical team with evidence based on previous performance but also provide context for making better decisions while the team navigates complex situations. Nonetheless, AI models supporting the decision-making of the surgical team will require interoperability and standardization with other sources of data available before surgery (101), and solutions to the challenges inherently faced in their regulation.

Implications for the surgical team

AI models introduced into the OR will change the way surgical teams work; however, their introduction will be subject to friction and barriers (Table 2). While AI models are likely to empower the team, the recording aspect of their performance may result in reluctance to be observed during surgery. This is not surprising, as AI models will be gradually introduced with the purpose of understanding workflows, quantifying performance, and ultimately allowing for task automation in the OR, with potentially serious repercussions for the surgical team.

AI models will also allow for the democratization of information available during surgery, bringing benefits to other members of the surgical team across different levels of expertise. However, the implications of such a trend are likely to lead to the degradation of skills, i.e., with fewer subject-matter experts and more generalists. AI methodology should target the empowerment of members of the surgical team beyond surgeons, allowing them to speak up by presenting data-driven evidence. Therefore, any mechanism that informs the entire team about what is currently happening and what can be anticipated from different surgical perspectives will add value.

Relationships among diverse AI models available in the OR that are trained for different purposes with data across the entire patient journey will provide an opportunity for the surgical team to improve surgical decision-making. For example, there will be models that can understand workflows from surgical videos and models that can integrate biomarkers resulting from tissue biopsy, histology, and gene sequencing. While there will be no need to manually write surgical notes to reflect what happened during surgery, the surgical team will require a mechanism to confirm that the outputs from these models are accurate and trustworthy.

New roles are anticipated to be seen in the OR in the near future. Data scientists are likely to work in the OR in tandem with the surgical team to record, curate, and investigate the performance and interpretability capabilities of AI models as they are introduced intraoperatively (102). System integrators will ensure that tools run reliably, replacing the roles currently performed by vendors of specific platforms introduced into the OR. Data in the context of interpretable AI model mechanisms will change the way surgical teams work. Whether these roles will result from upskilling current members of the surgical team through mandatory training or from hiring new staff will be subject to financial considerations at hospitals. Moreover, these potential roles will be more complex to perform, as they may involve processing vast amounts of data and interpreting models with diverse downstream tasks likely to be running simultaneously in the OR. It is unclear how surgical teams will be equipped with data science skills beyond recording and curating data. Any upskilling strategy for surgical teams is likely to be slow compared to the fast-paced technological progress that AI has experienced in recent years.

LLMs introduced in the OR will have an impact on the effectiveness of surgical teams and will revolutionize how they work, especially when designed for co-production. Designing tools with humans in the loop will be particularly important to minimize the risks and barriers related to decision-making support in the OR. When AI tools are deployed to improve surgical practices and patient outcomes in personalized surgery, surgical teams will need enough competency to interpret their outputs in order to build evidence to reliably optimize decisions.

Surgical robotics

Contemporary surgical robotic systems predominantly maintain a human-in-the-loop control paradigm, usually with limited autonomy (103). These systems typically employ a leader–follower architecture, whereby a surgeon manipulates instrument controls from a dedicated console, transmitting commands to a multi-arm robotic system, usually comprising multiple (e.g., three to four) rigid, electromechanically actuated arms. The da Vinci surgical system is the most established example of this (103, 104), offering precise endoscopic manipulation, motion scaling, tremor filtration, and, more recently, haptic force feedback. Systems such as the Bitrack robotic platform (Rob Surgical) (105) follow similar architectural principles. Parallel developments in modular platforms—such as Dexter (Distalmotion), Versius (CMR Surgical), and Hugo RAS (Medtronic)—are designed to decouple robotic arms from a centralized cart, enhancing flexibility in OR layout (106). The human-in-the-loop paradigm is also widely recognized for its potential to enable remote and geographically distributed surgery (107). Advances in networking, control strategies, and shared autonomy have further reinforced the feasibility of telesurgery, positioning human-supervised robotic platforms as a promising foundation for future remote surgical interventions.

In addition to laparoscopic applications, image-guided robotic systems are prevalent in orthopedic and neurosurgical procedures. These platforms integrate preoperative imaging, intraoperative navigation, and robotic guidance to increase positional accuracy under constraint-based guidance rather than improving dexterous manipulation. Examples include knee and hip arthroplasty (Mako; Stryker), spinal instrumentation (Mazor; Medtronic), and stereotactic neurosurgery (ROSA series; Zimmer Biomet). While remaining fundamentally human-in-the-loop, these platforms can support limited task-level automation, for example, through predefined trajectories and virtual safety boundaries.

Microsurgery is also a key domain of surgical robotics research, encompassing clinical applications where sub-millimeter precision is required. Examples include research systems in ophthalmology, with notable advances in the first-in-human evaluation of the Horizon system for cataract surgery (108) and the recent demonstration of autonomous subretinal and intravascular therapy delivery (109, 110). Similarly, extensive funding has been secured by commercial systems, including the Symani platform (Medical Microinstruments), the MUSA system (Microsure), and the PRECEYES Surgical System (PRECEYES). These systems focus on neurovascular anastomosis, lymphaticovenous anastomosis, and vitreoretinal surgery, respectively (111–113).

Other robotic technologies are increasingly being translated into clinical use. Continuum robotic manipulators, characterized by flexible, snake-like structures that can bend continuously along their length, and single-port access systems, which allow multiple instruments to be introduced through a single incision or trocar, enable improved dexterity and anatomical access through constrained entry points (114, 115). Other relevant robotic systems include micro- and nanorobotics, in which intervention is no longer mediated by articulated tools but by swarms of functional agents. These systems challenge classical definitions of surgical robotics by shifting the focus from the OR to the patient’s internal micro-environment. Regardless of the underlying technology or the stage of development, surgical robotic systems remain fundamentally human-in-the-loop, with limited task-level autonomy. They have primarily been used to augment precision, dexterity, or targeting, while perception, decision-making, and execution continue to rely on direct clinician control.

Although robotic systems have the potential to significantly redefine intraoperative workflows and team dynamics, integration remains challenging owing to the substantial space demands of current devices, the necessity for well-trained multidisciplinary teams, the difficulty of maintaining efficient movement and coordination in crowded ORs, and the time required to prepare the system between cases (116). Moreover, surgeons have had to acquire new competencies in robotic console operation, instrument calibration, and procedural planning within a robotic environment—skills that differ fundamentally from traditional open or laparoscopic techniques (117). Robotic surgery training programs now include virtual reality simulations, procedural dry lab training using synthetic models mimicking human tissues, and credentialing pathways tailored to specific platforms (118). In parallel, surgical assistants, scrub nurses, and circulating staff are required to develop proficiency in robotic docking, arm positioning, and maintaining sterile workflows within increasingly complex setups (119). Anesthesiologists must also adapt to the altered physical configurations and patient access constraints introduced by large robotic systems (116). Thus, the transition to robotic-assisted surgery has introduced a steep, multidisciplinary learning curve; at the same time, it has fostered more protocol-driven, collaborative team environments centered on technology.

Vision

Here, we outline our vision for the future application of surgical robotics across the preoperative, intraoperative, and postoperative phases, together with its implications for the surgical team (Figure 3).

Figure 3

Preoperative phase

The future surgical environment will incorporate robotic automation early in the workflow to optimize preparation and planning. Robotic systems will autonomously perform patient draping and post-draping registration (77) of both robotic and non-robotic components within the OR, significantly reducing setup time and alleviating the workload of scrub nurses. Advanced, robot-specific surgical planning and path optimization tools grounded in the best clinical practices will assist surgeons—particularly novices—by providing them with tailored operation and instrument path plans. These systems will account for all robotic and non-robotic elements within the OR, enabling real-time collision avoidance and enhancing safety. Fully robotic operating tables capable of precise patient positioning, such as the CyberKnife system (Accuray), will minimize the need for manual repositioning by the surgical team, thereby streamlining workflow and conserving human resources.

Intraoperative phase

During surgery, modular robotic units integrated seamlessly into the OR infrastructure—whether mounted on ceilings, embedded in floors, or deployed as mobile docking stations—will combine conventional articulated kinematics with advanced soft robotic or constant curvature (i.e., continuum robotic) end-effectors, facilitating dexterous and minimally invasive access to complex anatomical regions. Interchangeable, domain-specific tool sets will allow for flexible adaptation to a wide range of procedures. Handheld robotic instruments will extend surgeons’ precision and offer tremor-reduction capabilities (120, 121) while fostering greater clinician acceptance by delivering robotic functionality directly to surgeons’ hands. Compared with console-based systems, these devices offer enhanced proprioceptive feedback and increased agency, potentially lowering adoption barriers for clinical translation and preserving intuitive control.

Additionally, for emergency situations, robotic cardiopulmonary resuscitation (CPR) devices—such as the clinically established LUCAS system (Stryker)—already provide automated, high-quality chest compressions. Future iterations are anticipated to integrate real-time cardiac monitoring and autonomous defibrillation capabilities, enabling a fully closed-loop emergency response system to become a standard feature in any robotic OR. Their integration will be essential for reducing response time, maintaining patient and staff safety, and supporting the seamless execution of life-saving interventions in high-stakes contexts.

Assistive robotic units will transform intraoperative support roles. Robotic scrub assistants will autonomously manage instrument handoffs, tool tracking, and tray logistics, thereby redefining the traditional scrub nurse role into one focused on supervising robotic workflows. Similarly, the circulating nurse’s role will evolve into that of a coordinator managing fleets of mobile robotic platforms responsible for supply transport, restocking, and environmental control. Robotic imaging systems integrated within hybrid ORs will autonomously position and configure diagnostic modalities, optimizing spatial efficiency and procedural integration.

Postoperative phase

Postoperative workflows will benefit from enhanced robotic automation and maintenance protocols. Self-cleaning technologies, such as ultraviolet (UV)-light sterilization and coordinated robotic cleaning fleets, will improve sterilizability and reduce manual labor. This will enable teams to transition to roles that emphasize quality auditing and routine maintenance of robotic systems, ensuring operational readiness and sustained patient safety. These advances will free clinical personnel to concentrate more fully on patient care and recovery.

Implications for the surgical team

The progressive automation and integration of robotic systems will profoundly reshape clinical roles and team dynamics within the OR. New specialized positions are likely to emerge, including robotic systems operators who combine technical expertise with surgical knowledge to manage, calibrate, and troubleshoot complex platforms in real time. Robotics integration engineers will become indispensable for orchestrating interoperability across robotic subsystems and hospital information technology (IT) infrastructure, effectively transforming the OR into a cyber-physical ecosystem. The roles of scrub and circulating nurses will evolve: scrub nurses will supervise robotic scrub assistants and oversee workflow integration, while circulating nurses will coordinate autonomous logistics robots responsible for supply chain management and environmental control (83).

Although surgeons will continue as procedural leaders, they may increasingly divide their responsibilities between direct operation of handheld or console-based robotic systems and the orchestration of multi-robot workflows enhanced by image guidance. Anesthesiologists will adapt to limited patient access by overseeing increasingly comprehensive physiological monitoring systems and working closely with remote patient monitoring and emergency response units. This evolution will require interdisciplinary training that blends robotics, human–machine interaction, and clinical expertise, ultimately fostering a highly collaborative and technologically integrated surgical ecosystem.

Autonomy and human–robot interaction

The future of surgical robotics will be defined by a dynamic spectrum of autonomy (103)—ranging from direct, fully manual teleoperation to context-aware, AI-driven autonomy capable of executing complex procedures independently. Fully autonomous surgery has already been performed on ex vivo porcine gallbladder specimens without human intervention, demonstrating its capability for complex surgical interventions (71). Rather than aiming for blind automation from a human perspective (i.e., blackbox AI), technology is likely to center on gradual, task-specific autonomy where clinicians always retain overarching responsibility, decision-making, and supervisory control. Even at present, elements of task autonomy are becoming clinically viable, including automated suturing, wound closure, stapling (122), and collaborative actions such as camera/endoscope steering (123). These represent narrow, well-defined tasks that benefit from consistency and precision yet remain under human oversight. The question is how and when such delegation should occur once the surgical team recognizes that delegating a given task to an agent provides a clear benefit for addressing a capability gap (e.g., increased precision, improved comfort, or reduced variability) or delivering efficiency gains (e.g., reduced fatigue, procedure time, or exposure to hazards such as radiation). For instance, the CorPath GRX system (Siemens Healthineers) helps reduce operator radiation exposure and the need for prolonged use of heavy lead aprons during endovascular interventions, while the AquaBeam/Hydros system (PROCEPT BioRobotics) achieves greater precision in prostate tissue resection based on the surgeon’s defined treatment boundaries and targets in urology. These examples demonstrate that human-in-the-loop autonomy may first emerge in tasks where precision, consistency, and reduced occupational risk provide an immediate benefit, with the surgeon retaining responsibility for planning and intraoperative decision-making. As surgical robotics evolves, clinicians will increasingly be able to customize the degree of autonomy employed during procedures, selecting which steps to delegate and which to retain for manual execution based on personal experience, case complexity, and patient-specific factors. To achieve this, collaboration and gradual clinical translation will be key to supporting trust calibration—balancing under-reliance and automation bias—by aligning user reliance with demonstrated system reliability across contexts. This type of calibrated interaction is critical for safe deployment and underpins the progressive clinical translation of AI-assisted workflows.

Robotic perception and embodied AI

The path toward higher autonomy in robotic surgery requires robust robotic perception capabilities. Sensor technologies are central to this progression and include force and pressure sensors that enable safer tissue interaction (124), optical sensors that detect geometry and differentiate tissue types (125), and contactless proximity sensors (126) that allow for anticipatory motion planning. To support adaptive, context-sensitive autonomy, future surgical robots will increasingly rely on structured, computational representations of their bodies, surgical tools, patients’ anatomy, and the OR environment (Figure 4). These models are central to the concept of embodied AI, where intelligence emerges not just from abstract computation, but from physical interaction, sensory coupling, and real-world context. Unlike conventional robotic systems with rigid programming, embodied AI systems maintain situational awareness by integrating multimodal sensor data (e.g., vision, force, and proximity) into a unified spatial-temporal understanding. Techniques such as simultaneous localization and mapping (SLAM) (127) can be foundational in this domain, enabling robots to dynamically map and localize themselves within the complex OR geometry.

Figure 4

Crucially, the learned representation of the environment by world models (e.g., agents navigating patient-specific vasculature during mechanical thrombectomy) (67) enables computationally efficient long-term planning, leveraging human interaction (10) and leading to predictive and anticipatory behaviors, such as avoiding collisions with moving humans and other robots, recognizing task progression, and planning multiple steps. Additionally, single configurations of world models are able to generalize across multiple benchmark tasks, further supporting shared autonomy scenarios in which the robot can intelligently defer to the surgeon when uncertainty is high or step in to assist when precision or repeatability is paramount.

As robotic autonomy advances, a major enabler will be on-hardware learning, through which robots can learn from data collected during clinical operations directly on embedded systems or edge devices (42), rather than relying solely on cloud-based training pipelines. This allows for real-time model adaptation, privacy-preserving data handling, and low-latency decision-making, all of which are critical in surgical environments. For example, embodied surgical robots could incrementally refine their grasping strategies or path-planning behaviors over time to adapt to the specific preferences of different surgeons or the nuances of varied patient anatomies. These systems will not be “trained once, deployed forever”, but rather continuously evolve in situ.

Together, world models and embodied AI represent a paradigm shift: from reactive, pre-programmed machines toward causal (128), truly adaptive surgical collaborators capable of understanding, reasoning, and evolving within the complexity of real-world clinical environments.

Human-in-the-loop collaboration

Importantly, the integration of autonomy is not intended to replace clinicians, but rather to enhance their capabilities through close human–robot interaction, whereby robotic systems provide consistent, adaptive execution, while clinicians guide, oversee, and intervene as needed. This human-in-the-loop approach ensures safety, supports clinical decision-making, preserves human agency, and aligns with ethical and legal frameworks for medical responsibility. This strategy goes beyond transmitting knowledge, allowing the dynamic interplay between humans and AI models to achieve a shared objective. However, more elaborate designs and experiments are needed in the context of surgery, as recent evidence highlights that, on average, human–AI combinations perform significantly worse than the best of humans or AI alone, while only human augmentation is achieved when compared to a baseline related to humans alone (129). These findings show that over- and under-reliance on AI by humans has an effect on human–AI synergy and that tasks involving decision-making are the worst, while tasks involving content creation are beneficial.

To enable such a dynamic, close human–robot interaction, future surgical systems must support multimodal interaction capabilities that go beyond conventional physical interfaces. At the lowest level, spatial awareness supports collision avoidance and coordinated motion around clinicians and other robots, forming the foundation for safe coexistence in a crowded OR. At a higher level, systems will incorporate voice commands, restricted to privileged users to maintain procedural integrity and avoid accidental activation. Gesture recognition and gaze tracking can offer hands-free control of robotic elements, camera views, or instrument positioning, which is especially valuable during sterile procedures or in confined spaces. When combined, these modalities create a redundant and adaptive interaction architecture capable of adjusting to user preference, workload, and procedural context. To extend surgical reach beyond the immediate OR, next-generation systems will offer distributed user interface architectures accessible from inside the OR, adjacent control rooms, or even remote locations. Remote experts and trainees will be able to observe, annotate, or even interact with surgical workflows from a safe distance and in real time (42), opening new avenues for collaboration, tele-mentoring, and education.

Together, these multimodal and distributed interaction paradigms will serve as the human interface layer that underpins the safe, transparent, and adaptive application of surgical autonomy. Such autonomy must be aligned to collaborative conceptual frameworks to provide algorithms carefully designed for a wide range of human–AI engagement, from adaptive instrumentation to proactive assistance, joint inquiry, and co-production capabilities (130) (Figure 5). These frameworks form the technological basis for trustworthy human–robot collaboration, setting the stage for the next evolution of user interfaces and experience design in robotic surgery.

Figure 5

Implications for the surgical team

The progression toward graduated, task-specific autonomy in surgical robotics has significant implications for clinical roles in the OR. Surgeons will increasingly shift from being direct instrument operators to acting as supervisors and strategic decision-makers—selecting appropriate autonomy levels for individual tasks, overseeing robotic execution, and intervening when necessary. This requires new competencies, including familiarity with robotic system behavior, multimodal user interfaces (e.g., voice, gesture, and haptics), and real-time decision-making in collaboration with autonomous agents. The roles of scrub and circulating nurses will also evolve: scrub nurses may oversee the setup and monitoring of autonomous task sequences, while circulating nurses could support system-level coordination and ensure operational safety in proximity-based human–robot collaboration. As autonomy increases, new professional roles may emerge, such as intraoperative autonomy coordinators or clinical robotics technicians, responsible for monitoring workflows, calibrating sensors, and maintaining robotic infrastructure. Overall, the cognitive and procedural workload will be redistributed, reducing routine burdens while demanding higher-level oversight and interdisciplinary collaboration, particularly at the intersection of clinical expertise and robotic intelligence.

Ethics, health equity, and regulation

The integration of AI and robotics into surgery is not only a matter of technical innovation. These systems will reshape decision-making, healthcare access, and governance. Ethics, health equity, and regulation are therefore inseparable from technological progress, and addressing them is essential to ensuring safe, fair, and sustainable adoption of these technologies (Table 3, Figure 6).

Figure 6

Ethical considerations

Several ethical concerns arise when discussing the future of AI and robotics in the OR. Perhaps the most pertinent question regarding the evolving role of surgical teams is: who is ultimately in charge? Currently, the public is not ready for fully automated surgery (131). Future AI systems will almost certainly outperform human surgeons in predicting intraoperative outcomes, perhaps not due to some form of advanced autonomy but as a function of their sheer computational capacity and ability to analyze large, multimodal datasets in real time (132). However, AI is unlikely to be solely relied upon for decision-making for three key reasons. First, if humans maintain a role in the operating theater, liability for errors will be ambiguous (i.e., was it the fault of the human, the algorithm, or the manufacturer?) (133). Second, data-driven AI will struggle with complex, dynamic scenarios such as emergencies, outlier cases, or pioneering/first-in-human procedures (134). Third, surgeons generally have optimal access to and understanding of the operative context (135), and thus, decision-making authority in the OR is structured around them. If the very act of decision-making is assigned to AI, information may be immediately democratized, disrupting the established hierarchy of authority and team communication during critical events, at the same time requiring substantial expertise to evaluate an AI-based decision. Consequently, a human-in-the-loop solution will be favored, with AI models generating various options for the next step in a procedure, along with the corresponding probability and confidence interval for a positive outcome, for a surgeon to select, modify, and execute. Based on the surgeon’s decision, instructions are distributed to the rest of the team. This clarifies liability and narrows operational information access to the core decision-makers in the OR, ensuring a clear chain of command.

Bias is another key ethical concern. If left unaddressed, it risks exacerbating health inequalities in underrepresented populations (136). Bias typically originates in the training phase: models trained predominantly on a specific demographic will underperform on a generalized dataset (137). Careful curation of training datasets is, therefore, the first step in minimizing bias. Datasets should adequately represent all relevant demographics and actively oversample (through targeted data collection or synthetic data augmentation) underrepresented groups (138). Data should be pre-processed to remove or reweight connections to protected features such as race, gender, or age. Techniques such as adversarial debiasing, hyperparameter tuning, regularization, model drift detection, and retraining should be implemented (138). Federated learning is particularly promising as it enables decentralized institutions around the world to continuously update AI model parameters without sharing the raw data (82, 139). The possibility of bias should also trigger a discussion on whether AI models should be deployed globally or locally; while the prospect of a global, “one-size-fits-all” model is enticing, tailored models will usually outperform generalized models on a local population (140) and strategies such as federated learning can be difficult to implement due to logistical and quality realities (139). A more practical, transfer learning-based solution might be to take “off-the-shelf”, global AI models and refine these via domain-/task-adaptive pre-training and parameter fine-tuning to reflect local populations (141).

Moreover, the traditional model of informed consent does not adequately reflect the use of patient data to train AI software or the influence of AI systems on surgical decision-making. Consent should therefore address two distinct elements: the extent to which AI is involved in the procedure, and whether data may be reused for research or model development. Patients should be given personalized, non-technical, clear explanations of AI involvement and the associated risks (142). Where feasible, tiered consent should allow patients to agree to AI-assisted care while declining secondary uses of their data, such as algorithm training or commercial development. For complete transparency, patients should be informed about how their data are anonymized, the relevant retention period, and who the end user is. However, this information should be balanced against clinical practicability, and consent processes should avoid overwhelming patients with technical details. One strategy to accomplish this would be to move away from static, “instantaneous” consent forms to a continuous process that allows patients to provide, modify, or withdraw consent for specific data uses over the course of their care as the clinical context and data uses evolve.

Health equity

Different lenses have been considered in relation to data equity, including representation equity, feature equity, access equity, outcome equity, procedural equity, and decision-making equity to target all populations, ensure individualism, guarantee fair access, avoid unintended consequences, ensure fair AI chains, and guarantee transparency, respectively (143). Health equity may be improved via AI, but a growing reliance on AI and robotic systems in surgery also risks exacerbating health inequalities on a global scale (136, 144). Many low-resource settings lack the digital infrastructure, standardized processes, and trained personnel required to reliably capture, curate, and manage high-quality multimodal surgical data, resulting in datasets of poor quality and a systematic under-representation in model development. Paradoxically, these technologies remain unaffordable for the healthcare systems that would benefit the most (145). Stakeholders across academia and industry must champion frugal innovation by developing cost-effective solutions without compromising on outcomes (146). Collaborations with academic and surgical teams to build local expertise and address knowledge inequality should be encouraged (147). Low-resource nations are also likely to have inadequate regulatory and ethical frameworks to ensure the safe implementation of new technology. This can lead to the predatory practice of “ethics dumping”, in which firms trial healthcare products in these regions without adhering to their home country’s strict standards (148). Careful oversight is needed to support providers in these regions and create robust frameworks for implementation without exploitation.

Regulation

Regulators face the immense challenge of keeping pace with rapidly evolving AI and robotic technology in a field that has traditionally been risk-averse and slow to accept innovation. Crucially, they must ensure that all AI and robotic devices entering the market are safe for patient use, which will require a comprehensive revamp of compliance standards and regulatory frameworks that currently lag behind technological progress (149).

In the United States, the majority of Food and Drug Administration (FDA)-approved robots are classified as class II medical devices (moderate risk) (150). As these systems begin showcasing higher levels of autonomy, reclassification to class III, reflecting the increased risk, may become necessary. Currently, surgical robots are licensed by the FDA through either the 510(k) pre-market notification pathway (for devices with similar products or “predicates” already approved) or the de novo pathway (for entirely new products) (150). However, the 510(k) route allows for “predicate creep”, in which manufacturers add significant features to existing devices while claiming equivalence without the otherwise mandatory rigorous scrutiny: this gap is exploitable and poses a risk to patient safety (151). Indeed, a recent study found that recalls of FDA-cleared AI-enabled medical devices largely involved recently approved products with poor clinical validation, often produced by large, publicly traded companies through the 510(k) pathway (152). Legislation and guidance are needed to define permissible AI systems to protect patient safety and data privacy (153).

Regulatory approval is contingent on strong clinical evidence of equivalent or better outcomes, typically obtained through clinical trials. Trials for advanced AI and robotic systems should adopt standardized metrics to evaluate bias, accuracy, error, system reliability, vulnerability, costs, and impact on sustainability (154). Incorporating AI-specific reporting standards such as the CONSORT-AI (Consolidated Standards of Reporting Trials–Artificial Intelligence) and the SPIRIT-AI (Standard Protocol Items: Recommendations for Interventional Trials–Artificial Intelligence) should become routine practice (155, 156). Surgical robots should be developed and evaluated in accordance with the IDEAL (Idea, Development, Exploration, Assessment, and Long-term monitoring) framework (154). Trials should extend beyond patient outcomes and incorporate measures of team performance, such as human–machine interaction, usability, interpretability, and explainability (157). Regulators should also scrutinize surgical teams using these systems for procedures in which extensive trials have not taken place; the FDA has previously issued cautions to healthcare providers routinely offering robotic surgery for mastectomy and other oncological interventions (158), procedures that were unapproved and lacked evidence of equivalent survival and cancer-free outcomes.

Clinical trial methodologies will also need to evolve in the age of AI. Randomized controlled trials (RCTs) will use AI to optimize the study protocol, design, randomization, and data collection; detect deviations from the protocol/expected outcomes; and determine ideal endpoints (159, 160). Simulation will play a significant role in the future; the scope of in silico trials will expand as modeling becomes increasingly sophisticated (161). Prospective RCTs may be replaced in part by AI-driven retrospective studies that examine observational data, use propensity score matching (162), create virtual controls to estimate outcomes for patients who did not receive the specified intervention (163), and use a wide range of data from patient records (including imaging from pathology and radiology reports and written ward and clinic notes) to predict individual outcomes.

Standardization should not only encompass reporting metrics but also broad regulatory practices. Global collaboration is key; a promising example is the collaboration between the FDA, the United Kingdom’s Medicines and Healthcare Products Regulatory Agency (MHRA), and Health Canada to harmonize the regulatory framework for the development of AI and AI-assisted medical devices (164, 165).

Regulators should ensure adequate surgical team training during AI and robotic system deployment. Although professional bodies (e.g., the Royal Colleges in the United Kingdom) set the curriculum for surgical training, the shift toward data-informed practice will require careful oversight and regulatory input. The change in required competencies should be reflected at all points in the training pathway; medical schools, for instance, should begin teaching statistics, mathematical models of probability, and the fundamentals of data science (166–168). Simulation training involving entire surgical teams should be carried out to ensure smooth communication and interaction with AI models. Surgeon de-skilling must be proactively addressed to ensure baseline competencies in manual surgical skills are maintained.

Robust post-market monitoring becomes critical after deployment. Existing device adverse event databases, such as the FDA’s Manufacturer and User Facility Device Experience (MAUDE) and the MHRA’s yellow card scheme, have been criticized for inadequately capturing AI-specific failures (169). AI-specific taxonomies with reporting categories such as model drift, algorithmic stability, and security vulnerabilities need to be urgently integrated. Surgical teams should be empowered to report these failures, and transparency should be emphasized.

As mentioned earlier, compliance standards lag behind the pace of progress in surgical AI and robotics. This gap is likely to widen—for instance, the current practice is to regard the majority of soft tissue surgical robots as medical electrical devices rather than true robots (since they are not programmable and cannot carry out autonomous activity) (170). As a result, these systems are held to the same International Organization for Standardization (ISO) and International Electrotechnical Commission (IEC) standards applied to other medical devices, such as ISO 14971 for risk stratification, ISO 13485 for quality management systems, and IEC 60601 for electrical components. This practice does not reflect the differing risk profile or the trend toward autonomy in surgical robots. Promising steps have been made, including the introduction of IEC 80601-2-77 within IEC 60601 (170), which specifically targets surgical robots (171), but more work is required to ensure that these standards remain relevant in the future.

We are likely to see the increasingly widespread use of active learning and adaptive AI that tailors its parameters based on new inputs. This represents a unique regulatory challenge, as the system in question can alter its decision-making process over a period of time. The vast majority of FDA-approved AI software ‘locks’ weights and requires re-review if these parameters are altered (172, 173). Devices using adaptive AI need standardized protocols detailing updates, validation testing, and patient populations (174). Regulators should perform spot checks to ensure that model performance is maintained. Transparent audit trails and automated diagnostic checks to ensure performance drift is recognized early should be mandatory.

Conclusion

The integration of AI and robotics into surgery represents a revolutionary opportunity to enhance the performance, situational awareness, decision-making, and effectiveness of surgical teams. Future surgical AI will use multimodal data to provide real-time intraoperative guidance and decision support, automated documentation, and the quantification of surgical performance. Leveraging causal AI will enable the identification of true cause-and-effect relationships, allowing surgeons to accurately visualize the outcomes of a variety of actions and thus improve results. On the robotics front, current systems will evolve to exhibit greater degrees of autonomy while maintaining human-in-the-loop control. Surgical robots will increasingly utilize embodied AI with advanced sensor integration to enhance spatial perception. Robotic integration in the OR will not just be limited to surgical robots—it will extend to include assistive devices for instrument handling, patient positioning, sterilization support, and emergency interventions. However, the safe deployment of these systems requires overcoming significant ethical and regulatory challenges. Human surgeons must continue to be the primary decision-makers, and insights from an AI model should be presented differently to members of the surgical team based on their roles, to maintain a clear chain of authority. Academia and industry should develop new modes of collaboration with healthcare systems in lower-income nations, working together to build cost-effective AI and robotic ecosystems. On the regulatory side, current licensing pathways, device classification norms, post-market monitoring databases, and compliance standards should be re-examined to reflect the increased risk profile of advanced AI and surgical robots. Clinical trials should adopt standardized metrics for the evaluation and reporting of AI software and should incorporate measures of human–AI and human–robot interaction. Regulators should oversee surgical training alongside professional bodies as practice transitions from being clinically driven to data driven. This complete surgical AI and robotics ecosystem will redefine the roles and makeup of surgical teams, shifting from minutely performative to globally supervisory functions and introducing new team members who will focus on managing the technical and data science framework of the future OR.

Factors indicating which surgical interventions are most likely to benefit from AI and robotics in the next few years include research volume, limited access, data availability, training complexity, real-time data analysis, critical time-to-treatment metrics, and the volume of unmet global surgical interventions, among others. We have identified four categories of surgeries that will benefit from technical development: (i) surgeries with short training pipelines, a large research base, and large data availability (e.g., minimally invasive and laparoscopic interventions); (ii) surgeries that are prone to robotic deployment, supported by advances in medical imaging (e.g., orthopedics and endoscopic surgery); (iii) surgeries with high training barriers for surgeons and high ethical and safety requirements (e.g., neurosurgery); and (iv) surgeries requiring higher precision and improved accuracy but that rely on large amounts of real-time data (e.g., oncology surgery, vascular and endovascular interventions, and microsurgery). Although the future course of innovation in AI and robotics and its effects on surgical teams are difficult to predict, our vision—grounded in key trends and challenges—seeks to offer stakeholders a structured framework to guide discussions about the responsible deployment of future surgical technologies. Ultimately, we must ensure, in a responsible and ethical manner, that healthcare professionals and patients benefit from AI and robotics to sustain, rather than disrupt, practice, refining the skills of care providers to achieve true personalized surgery and propel procedural and technological advancements.

References

• 1

  ShrimeMGBicklerSWAlkireBCMockC. Global burden of surgical disease: an estimation from the provider perspective. Lancet Global Health (2015) 3:S8–9. doi: PIIS2214-109X(14)70384-5 doi: 10.1016/S2214-109X(14)70384-5

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 2

  RoseJChangDCWeiserTGKassebaumNJBicklerSW. The role of surgery in global health: analysis of United States inpatient procedure frequency by condition using the Global Burden of Disease 2010 framework. PloS One (2014) 9(2):e89693. doi: 10.1371/journal.pone.0089693

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 3

  MearaJGLeatherAJMHaganderLAlkireBCAlonsoNAmehEAA. Global Surgery 2030: evidence and solutions for achieving health, welfare, and economic development. Lancet (2015) 386(9993):569–624. doi: 10.1016/S0140-6736(15)60160-X

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 4

  RoseJWeiserTGHiderPWilsonLGruenRLBicklerSW. Estimated need for surgery worldwide based on prevalence of diseases: a modelling strategy for the WHO Global Health Estimate. Lancet Glob Health (2015) 3 Suppl 2(Suppl 2):S13–20. doi: 10.1016/S2214-109X(15)70087-2

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 5

  PereraSKJacobSWilsonBEFerlayJBrayFSullivanRet al. Global demand for cancer surgery and an estimate of the optimal surgical and anaesthesia workforce between 2018 and 2040: a population-based modelling study. Lancet Oncol (2021) 22(2):182–9. doi: 10.1016/S1470-2045(20)30675-6

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 6

  NepogodievDMartinJBiccardBMakupeABhanguA. Global burden of postoperative death. Lancet (2019) 393(10170):401. doi: 10.1016/S0140-6736(18)33139-8

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 7

  RanganathanPDareAHarrisonEMKinghamTPMutebiMParhamGet al. Inequities in global cancer surgery: challenges and solutions. J Surg Oncol (2024) 129(1):150–8. doi: 10.1002/jso.27551

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 8

  D’EttorreCMarianiAStilliARodriguez y BaenaFValdastriPDeguetAet al. Accelerating surgical robotics research: a review of 10 years with the da Vinci research kit. IEEE Robot Automat Magazine (2021) 28(4):56–78. doi: 10.1109/MRA.2021.3101646

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 9

  SemeraroFGriffithsACangelosiA. Human–robot collaboration and machine learning: a systematic review of recent research. Robot Comput-Integr Manuf (2023) 79. doi: 10.1016/j.rcim.2022.102432

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 10

  LongYWeiWHuangTWangYDouQ. Human-in-the-loop embodied intelligence with interactive simulation environment for surgical robot learning. IEEE Robot Automat Lett (2023) 8(8):4441–8. doi: 10.1109/LRA.2023.3284380

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 11

  Maier-HeinLEisenmannMSarikayaDMärzKCollinsTMalpaniAet al. Surgical data science – from concepts toward clinical translation. Med Image Anal (2022) 76:102306. doi: 10.1016/j.media.2021.102306

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 12

  RiskinDJLongakerMTGertnerMKrummelTM. Innovation in surgery: a historical perspective. Ann Surg (2006) 244(5):686–93. doi: 10.1097/01.sla.0000242706.91771.ce

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 13

  KneeboneRWoodsA. Bringing surgical history to life. BMJ (2012) 345:e8135. doi: 10.1136/bmj.e8135

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 14

  FramptonSKneeboneR. The Lancet: an archive of surgical history. Lancet (2023) 402(10409):1222–4. doi: 10.1016/S0140-6736(23)02192-X

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 15

  AvelingELStoneJSundtTWrightCGinoFSingerS. Factors influencing team behaviors in surgery: a qualitative study to inform teamwork interventions. Ann Thorac Surg (2018) 106(1):115–20. doi: 10.1016/j.athoracsur.2017.12.045

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 16

  MakaryMASextonJBFreischlagJAHolzmüllerCGMillmanEARowenLet al. Operating room teamwork among physicians and nurses: teamwork in the eye of the beholder. J Am Coll Surgeons (2006) 202(5):746–52. doi: 10.1016/j.jamcollsurg.2006.01.017

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 17

  SteinlGKWeaverML. A little less conversation a little more action, please. Am J Surg (2024) 35:115802. doi: 10.1016/j.amjsurg.2024.115802

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 18

  StoneJLAvelingELFreanMShieldsMCWrightCGinoFet al. Effective leadership of surgical teams: a mixed methods study of surgeon behaviors and functions. Ann Thorac Surg. (2017) 104(2):530–7. doi: 10.1016/j.athoracsur.2017.01.021

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 19

  SunRMarshallDCSykesMCMaruthappuMShalhoubJ. The impact of improving teamwork on patient outcomes in surgery: a systematic review. Int J Surg (2018) 53:171–7. doi: 10.1016/j.ijsu.2018.03.044

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 20

  TørringBGittellJHLaursenMRasmussenBSSørensenEE. Communication and relationship dynamics in surgical teams in the operating room: an ethnographic study. BMC Health Serv Res (2019) 19:528. doi: 10.1186/s12913-019-4362-0

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 21

  McCullochPRathboneJCatchpoleK. Interventions to improve teamwork and communications among healthcare staff. Br J Surg (2011) 98(4):469–79. doi: 10.1002/bjs.7434

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 22

  BoetSLarriganSMartinLLiuHSullivanKJEtheringtonC. Measuring non-technical skills of anaesthesiologists in the operating room: a systematic review of assessment tools and their measurement properties. Br J Anaesthesia (2018) 121(6):1218–26. doi: 10.1016/j.bja.2018.07.028

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 23

  LiuYBoelsMGarcia-Peraza-HerreraLCVercauterenTDasguptaPGranadosAet al. LoViT: Long Video Transformer for surgical phase recognition. Med Image Anal (2025) 99:103366. doi: 10.1016/j.media.2024.103366

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 24

  BoelsMLiuYDasguptaPGranadosAOurselinS. SWAG: long-term surgical workflow prediction with generative-based anticipation. Int J CARS (2025). 20(2269–79) doi: 10.1007/s11548-025-03452-8

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 25

  AbercrombieJ. General surgery: GIRFT programme national specialty report. Getting It Right First Time (2017). Available at: https://gettingitrightfirsttime.co.uk/wp-content/uploads/2018/08/GIRFT-GeneralSurgery-Aug17-O1.pdf

  • Google Scholar

• 26

  KrummelTM. Forecasting innovation in surgery. Ann Surg (2014) 260(2):212–3. doi: 10.1097/SLA.0000000000000804

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 27

  KneeboneR. The vanishing art of doing. BMJ (2019) 364:k5326. doi: 10.1136/bmj.k5326

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 28

  ClaxtonG. Intelligence in the flesh: why your mind needs your body much more than it thinks (1st edition). New Haven, CT: Yale University Press (2016).

  • Google Scholar

• 29

  HaynesABWeiserTGBerryWRLipsitzSRBreizatA-HSDellingerEPet al. A surgical safety checklist to reduce morbidity and mortality in a global population. New Engl J Med (2009). 360:491–9. doi: 10.1056/NEJMsa0810119

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 30

  VercauterenTUnberathMPadoyNNavabN. CAI4CAI: the rise of contextual artificial intelligence in computer assisted interventions. Proc IEEE Inst Electr Electron Eng (2020) 108(1):198–214. doi: 10.1109/JPROC.2019.2946993

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 31

  RobertshawHKarstensenLJacksonBSadatiHRhodeKOurselinSet al. Artificial intelligence in the autonomous navigation of endovascular interventions: a systematic review. Front Hum Neurosci (2023) 17. doi: 10.3389/fnhum.2023.1239374

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 32

  VargheseCHarrisonEMO’GradyGTopolEJ. Artificial intelligence in surgery. Nat Med (2024) 30:1257–68. doi: 10.1038/s41591-024-02970-3

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 33

  FuchtmannJRiedelTBerletMJellAWegenerLWagnerLet al. Audio-based event detection in the operating room. Int J CARS (2024) 19:2381–7. doi: 10.1007/s11548-024-03211-1

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 34

  BoalMWEAnastasiouDTesfaiFGhamrawiWMazomenosECurtisNet al. Evaluation of objective tools and artificial intelligence in robotic surgery technical skills assessment: a systematic review. Br J Surg (2024) 111(1). doi: 10.1093/bjs/znad331

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 35

  ZhouTChaJSGonzalezGWachsJPSundaramCPYuD. Multimodal physiological signals for workload prediction in robot-assisted surgery. J Hum-Robot Interact (2020) 9(2):1–26. doi: 10.1145/3368589

  • CrossRef
  • Google Scholar

• 36

  DemirKCSchieberHWeiseTRothDMaierAYangSH. Deep learning in surgical workflow analysis: a review of phase and step recognition. IEEE J Biomed Health Inf (2023) 27(11):5405–17. doi: 10.1109/JBHI.2023.3311628

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 37

  LiuYHuoJPengJSparksRDasguptaPGranadosAet al. SKiT: a fast key information video transformer for online surgical phase recognition. IEEE/CVF In. Conf Comput Vis (2023) 21017–27. doi: 10.1109/ICCV51070.2023.01927

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 38

  MaRRamaswamyAXuJTrinhLKiyassehDChuTNet al. Surgical gestures as a method to quantify surgical performance and predict patient outcomes. NPJ Digit Med (2022) 5:187. doi: 10.1038/s41746-022-00738-y

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 39

  PsychogyiosDColleoniEAmsterdamBVLiC-YHuangS-YLiYet al. SAR-RARP50: segmentation of surgical instrumentation and action recognition on robot-assisted radical prostatectomy challenge. arXiv [preprint] (2023). doi: 10.48550/arXiv.2401.00496

  • CrossRef
  • Google Scholar

• 40

  MadaniANamaziBAltieriMSHashimotoDARiveraAMPucherPHet al. Artificial intelligence for intraoperative guidance: using semantic segmentation to identify surgical anatomy during laparoscopic cholecystectomy. Ann Surg (2022) 276(2):363–9. doi: 10.1097/SLA.0000000000004594

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 41

  LiYWuC-YFanHMangalamKBoXMalikJet al. MViTv2: improved multiscale vision transformers for classification and detection. In: Proceedings of the 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition. New York, NY: IEEE (2022). doi: 10.1109/CVPR52688.2022.00476

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 42

  QuarezJLiYIrzanHElliotMMacCormacOKnightJet al. MUTUAL: towards holistic sensing and inference in the operating room. Med Image Comput Comput Assist Interv (2025) 17:178–88. doi: 10.1007/978-3-031-77610-6_17

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 43

  GeratsBGAWolterinkJMBroedersIAMJ. 3D human pose estimation in multi-view operating room videos using differentiable camera projections. Comput. Methods Biomech. Biomed. Engin. Imaging Vis. (2023) 11(4):1197–205. doi: 10.1080/21681163.2022.2155580

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 44

  SrivastavVGangiAPadoyN. Unsupervised domain adaptation for clinician pose estimation and instance segmentation in the operating room. Med Image Anal (2022) 80:102525. doi: 10.1016/j.media.2022.102525

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 45

  SeenivasanLIslam M KannanGRenH. SurgicalGPT: end-to-end language-vision GPT for visual question answering in surgery. MICCAI (2023) 281–90. doi: 10.1007/978-3-031-43996-4_27

  • CrossRef
  • Google Scholar

• 46

  HeRXuMDasAKhanDZBanoSMarcusHJet al. PitVQA: image-grounded text embedding LLM for visual question answering in pituitary surgery. Med Image Comput Comput Assist Intervent (2024) LNCS 15006:488–98. doi: 10.1007/978-3-031-72089-5_46

  • CrossRef
  • Google Scholar

• 47

  YuanKSrivastavVYuTLavanchyJLMarescauxJMascagniPet al. Learning multi-modal representations by watching hundreds of surgical video lectures. Med Image Anal (2025) 105:103644. doi: 10.1016/j.media.2025.103644

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 48

  ZengZZhuoZJiaXZhangEWuJZhangJet al. SurgVLM: a large vision-language model and systematic evaluation benchmark for surgical intelligence. arXiv [preprint] (2025). doi: 10.48550/arXiv.2506.02555

  • CrossRef
  • Google Scholar

• 49

  ChaJSAthanasiadisDIPengYWuDAntonNEStefanidisDet al. Objective nontechnical skills measurement using sensor-based behavior metrics in surgical teams. Hum Factors (2024) 66(3):729–43. doi: 10.1177/00187208221101292

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 50

  JiangBQuarezJLiYIrzanHSaeedMElliotMet al. SphenoidTalk: towards real-time expert commentary prediction during transsphenoidal surgery [article 61]. In Proceedings of HSMR25: The 17th Hamlyn Symposium on Medical Robotics (2025). doi: 10.31256/HSMR25.31

  • CrossRef
  • Google Scholar

• 51

  WilsonCChahineSCristanchoSAquilSMandurahMLevineMet al. Unusual suspects: real-time physiological evaluation of stressors during laparoscopic donor nephrectomy. Can Urol Assoc J (2021) 15(4):E205–9. doi: 10.5489/cuaj.6647

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 52

  ProtserovSHunterJZhangHMashouriPMasinoCBrudnoMet al. Development, deployment and scaling of operating room-ready artificial intelligence for real-time surgical decision support. NPJ Digit Med (2024) 7:231. doi: 10.1038/s41746-024-01225-2

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 53

  GeorgenthumHCosentinoCMarozzoFLiòP. Enhancing surgical documentation through multimodal visual-temporal transformers and generative AI. arXiv [preprint] (2025). doi: 10.48550/arXiv.2504.19918

  • CrossRef
  • Google Scholar

• 54

  WagnerLJourdanSMayerLMüllerCBernhardLKolbSet al. Robotic scrub nurse to anticipate surgical instruments based on real-time laparoscopic video analysis. Nat Commun Med (2024) 5:156. doi: 10.1038/s43856-024-00581-0

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 55

  KhanDZValetopoulouADasAHanrahanJGWilliamsSCBanoSet al. Artificial intelligence assisted operative anatomy recognition in endoscopic pituitary surgery. NPJ Digit Med (2024) 7:314. doi: 10.1038/s41746-024-01273-8

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 56

  MohammadiIFirouzabadiSRHosseinpourMAkhlaghpasandMHajikarimlooBTavanaeiRet al. Predictive ability of hypotension prediction index and machine learning methods in intraoperative hypotension: a systematic review and meta-analysis. J Transl Med (2024) 22:725. doi: 10.1186/s12967-024-05481-4

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 57

  GiriRFirdhosSHVidaTA. Artificial intelligence in anesthesia: enhancing precision, safety, and global access through data-driven systems. J Clin Med (2025) 14(19):6900. doi: 10.3390/jcm14196900

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 58

  GranadosAHanYLucenaOVakhariaVRodionovRVosSBet al. Patient-specific prediction of SEEG electrode bending for stereotactic neurosurgical planning. Int J CARS (2021) 16:789–98. doi: 10.1007/s11548-021-02347-8

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 59

  BahlkeFEnglertFPopaMBourierFReentsTLennerzCet al. First clinical data on artificial intelligence-guided catheter ablation in long-standing persistent atrial fibrillation. J Cardiovasc Electrophysiol (2024) 35:406–14. doi: 10.1111/jce.16184

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 60

  PengJFioreGLiuYEllumKDasguptaDAshkanKet al. From pre- to intra-operative MRI: predicting brain shift in temporal lobe resection for epilepsy surgery. arXiv [preprint] (2026). doi: 10.48550/arXiv.2602.03785

  • CrossRef
  • Google Scholar

• 61

  ZbindenLNelsonNChenJ-TChenXKimJWAzizianMet al. Cosmos-Surg-dVRK: world foundation model-based automated online evaluation of surgical robot policy learning. IEEE Robot Autom Lett (2026) 11(5):5978–85. doi: 10.1109/LRA.2026.3675962

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 62

  BoelsMRobertshawHBoothTCGranadosADasguptaPOurselinS. Surgical robot learning: from demonstration and simulation to world models–a review. TechRxiv [preprint] (2025). doi: 10.36227/techrxiv.175691283.37220268/v1

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 63

  TrovatoGShikanaiMUkawaGKinoshitaJMuraiNLeeJWet al. Development of a colon endoscope robot that adjusts its locomotion through the use of reinforcement learning. Int J CARS (2010) 5:317–25. doi: 10.1007/s11548-010-0481-0

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 64

  PoreAFinocchiaroMDall’AlbaDHernansanzACiutiGArezzoA. Colonoscopy navigation using end-to-end deep visuomotor control: a user study. Proceedings of the 2022 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 23-27 Oct. 2022 (2022) 9582–88. doi: 10.1109/IROS47612.2022.9981480

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 65

  LazoJFLaiC-FMocciaSRosaBCatellaniMde MathelinMet al. Autonomous intraluminal navigation of a soft robot using deep-learning-based visual servoing. IEEE IROS (2022) doi: 10.1109/IROS47612.2022.9982141

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 66

  KarstensenLRitterJHatzlJErnstFLangejürgenJUhlCet al. Recurrent neural networks for generalization towards the vessel geometry in autonomous endovascular guidewire navigation in the aortic arch. Int J CARS (2023) 18:1735–44. doi: 10.1007/s11548-023-02938-7

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 67

  RobertshawHWuH-RGranadosABoothT. World model for AI autonomous navigation in mechanical thrombectomy. In: GeeJCAlexanderDCHongJIglesiasJESudreCHVenkataramanAet al., editors. Proceedings of the Medical Image Computing and Computer Assisted Intervention – MICCAI 2025. 28th International Conference, Daejeon, South Korea, September 23–27, 2025, Part IX. Cham: Springer (2026). 680–90. doi: 10.1007/978-3-032-05114-1_65

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 68

  RobertshawHJacksonBWangJSadatiHKarstensenLGranadosAet al. Reinforcement learning for safe autonomous two-device navigation of cerebral vessels in mechanical thrombectomy. Int J CARS (2025) 20:1077–86. doi: 10.1007/s11548-025-03339-8

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 69

  RobertshawHKarstensenLJacksonBGranadosABoothTC. Autonomous navigation of catheters and guidewires in mechanical thrombectomy using inverse reinforcement learning. Int J CARS. (2024) 19:1569–78. doi: 10.1007/s11548-024-03208-w

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 70

  SaeidiHOpfermannJDKamMWeiSLeonardSHsiehMHet al. Autonomous robotic laparoscopic surgery for intestinal anastomosis. Sci Robot (2022) 7(62):eabj2908. doi: 10.1126/scirobotics.abj2908

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 71

  KimJWChenJ-THansenPShiLXGoldenbergASchmidgallSet al. SRT-H: a hierarchical framework for autonomous surgery via language-conditioned imitation learning. Sci Robot (2025) 10:eadt5254. doi: 10.1126/scirobotics.adt5254

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 72

  ErgenAÇabukBYıldırımPYenerMDKizirSÇaklılıMet al. Design and use of assistant robotic arm in endoscopic transnasal surgery. Neurosurg Focus (2024) 57(6):E6. doi: 10.3171/2024.9.FOCUS24426

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 73

  NwoyeCIPadoyN. SurgiTrack: fine-grained multi-class multi-tool tracking in surgical videos. Med Image Anal (2025) 101:103438. doi: 10.1016/j.media.2024.103438

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 74

  QuarezJModatMOurselinSShapeyJGranadosA. ReCAP: recursive cross attention network for pseudo-label generation in robotic surgical skill assessment. In: QiuJWuJLanglotzCHuangBLeiZWuHet al., editors. AI for Clinical Applications. First International Workshops, Agentic AI 2025, CREATE 2025, and Clinical MLLMs 2025, Held in Conjunction with MICCAI 2025, Daejeon, South Korea, September 23 and 27, 2025, Proceedings. Cham: Springer (2025). 156–66. doi: 10.1007/978-3-032-06004-4_16

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 75

  ZhouTWachsJP. Needle in a haystack: interactive surgical instrument recognition through perception and manipulation. Robot Autonomous Syst (2017) 97:182–92. doi: 10.1016/j.robot.2017.08.013

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 76

  DaveCLiuWMEPearseCQuarezJLiYChallacombeBet al. Hands on: a comparative analysis of left- and right-hand workspaces in surgical robotic skills training. In: The Hamlyn Symposium of Medical Robotics (2025).

  • Google Scholar

• 77

  HuberMCavalcantiNADavoodiALiRMowerCECarrilloFet al. Localising under the drape: proprioception in the era of distributed surgical robotic system. arXiv [preprint] (2025). doi: 10.48550/arXiv.2510.23512

  • CrossRef
  • Google Scholar

• 78

  LiXMatsumotoNPasupuletyUDeoAYangCMoranJet al. End to end AI system for surgical gesture sequence recognition and clinical outcome prediction. arXiv [preprint] (2025). doi: 10.48550/arXiv.2511.11899

  • CrossRef
  • Google Scholar

• 79

  CiabattiniMCaladoASawantRKhanNPrezziDAntonelliMet al. Effect of nerve sparing gestures on positive surgical margins during robotic-assisted radical prostatectomy. In: Proceedings of HSMR25: The 17th Hamlyn Symposium on Medical Robotics (2025) doi: 10.31256/HSMR25.18

  • CrossRef
  • Google Scholar

• 80

  MascagniPAlapattDMuraliAVardazaryanAGarciaAOkamotoNet al. Endoscapes, a critical view of safety and surgical scene segmentation dataset for laparoscopic cholecystectomy. Sci Data (2025) 12:331. doi: 10.1038/s41597-025-04642-4

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 81

  SharmaSVannucciMLegoriLPScagliaMLaraccaGGMutterDet al. Early operative difficulty assessment in laparoscopic cholecystectomy via snapshot-centric video analysis. Int J CARS (2025) 20:1185–93. doi: 10.1007/s11548-025-03372-7

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 82

  LiYKunduSSBoelsMMahmoodiTOurselinSVercauterenT. UltraFlwr - an efficient federated medical and surgical object detection framework. arXiv [preprint] (2025). doi: 10.48550/arXiv.2503.15161

  • CrossRef
  • Google Scholar

• 83

  BelliniVRussoMDomenichettiTPanizziMAllaiSBignamiEG. Artificial intelligence in operating theatre management. J Med Syst (2024) 48:19. doi: 10.1007/s10916-024-02038-2

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 84

  DydaAPurcellMCurtisSFieldEPillaiPRicardoKet al. Differential privacy for public health data: an innovative tool to optimize information sharing while protecting data confidentiality. Patterns (2021) 2(12):100366. doi: 10.1016/j.patter.2021.100366

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 85

  JuroschFKrogerNKolbSMehmetiFMartensESpeidelSet al. 6G networks for the operating room of the future. Prog Biomed Eng (2024) 6:043001. doi: 10.1088/2516-1091/ad819c

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 86

  KhannaAWolfTFrankIKruegerAShahPSharmaVet al. Enhancing accuracy of operative reports with automated artificial intelligence analysis of surgical video. J Am Coll Surgeons (2025) 240(5):739–46. doi: 10.1097/XCS.0000000000001352

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 87

  HungAJChenJGillIS. Automated performance metrics and machine learning algorithms to measure surgeon performance and anticipate clinical outcomes in robotic surgery. JAMA Surg (2018) 153(8):770–1. doi: 10.1001/jamasurg.2018.1512

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 88

  EpplerMBSayeghASMaasMVenkatAHemalSDesaiMMet al. Automated capture of intraoperative adverse events using artificial intelligence: a systematic review and meta-analysis. J Clin Med (2023) 12(4):1687. doi: 10.3390/jcm12041687

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 89

  GlymourCZhangKSpritesP. Review of causal discovery methods based on graphical models. Front Genet (2019) 10:524. doi: 10.3389/fgene.2019.00524

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 90

  ScholkopfBLocatelloFBauerSKeNRKalchbrennerNGoyalAet al. Towards causal representation learning. Proceedings of the IEEE (2021) 109(5):612–34. doi: 10.1109/JPROC.2021.3058954

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 91

  ScholkopfBKugelgenJ. From statistical to causal learning. arXiv [preprint] (2022). doi: 10.48550/arXiv.2204.00607

  • CrossRef
  • Google Scholar

• 92

  OhNChoiG-SLeeWY. ChatGPT goes to the operating theatre: evaluating GPT-4 performance and its potential in surgical education and training in the era of large language models. Ann Surg Treat Res (2023) 104(5):269–73. doi: 10.4174/astr.2023.104.5.269

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 93

  AlhurA. Redefining healthcare with artificial intelligence (AI): the contributions of ChatGPT, Gemini, and Co-pilot. Cureus. (2024) 16(4):e57795. doi: 10.7759/cureus.57795

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 94

  World Health Organization. Ethics and governance of artificial intelligence for health: guidance on large multi-modal models. Geneva: WHO (2024). Available at: https://iris.who.int/bitstream/handle/10665/375579/9789240084759-eng.pdf?sequence=1

  • Google Scholar

• 95

  OngCSObeyNTZhengYCohanASchneiderEB. SurgeryLLM: a retrieval-augmented generation large language model framework for surgical decision support and workflow enhancement. NPJ Digit Med (2024) 7:364. doi: 10.1038/s41746-024-01391-3

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 96

  OzsoyEPellegriniCCzempielTTristramFYuanKBani-HarouniDet al. MM-OR: a large multimodal operating room dataset for semantic understanding of high-intensity surgical environments. Proceedings of the 2025 Conference on Computer Vision and Pattern Recognition (CVPR), 10-17 June 2025 (2025). 19378–89. doi: 10.1109/CVPR52734.2025.01805

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 97

  SchmidgallSChoJZakkaCHiesingerW. GP-VLS: a general-purpose vision language model for surgery. arXiv [preprint] (2024). doi: 10.48550/arXiv.2407.19305

  • CrossRef
  • Google Scholar

• 98

  LowCHWangZZhangTZhuoZZengZMazomenosEBet al. SurgRAW: multi-agent workflow with chain-of-thought reasoning for robotic surgical video analysis. IEEE Robot Autom Lett (2026) 11(4):4857–64. doi: 10.1109/LRA.2026.3665443

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 99

  LiSWangJDaiRMaWNgWYHuYet al. Robonurse-VLA: robotic scrub nurse system based on vision-language-action model. In: Proceedings of the 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), Hangzhou, China (2025). 3986–93. doi: 10.1109/IROS60139.2025.11246030

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 100

  NgCKBaiLWangGWangYGaoHYuanKet al. EndoVLA: dual-phase vision-language-action model for autonomous tracking in endoscopy. arXiv [preprint] (2025). doi: 10.48550/arXiv.2505.15206

  • CrossRef
  • Google Scholar

• 101

  LoftusTJTighePJFilibertoACEfronPABrakenridgeSCMohrAMet al. Artificial intelligence and surgical decision-making. JAMA Surg (2020) . 155(2):148–58. doi: 10.1001/jamasurg.2019.4917

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 102

  HashimotoDARosmanGRusDMeirelesOR. Artificial intelligence in surgery: promises and perils. Ann Surg (2018) 268(1):70–6. doi: 10.1097/SLA.0000000000002693

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 103

  AttanasioAScaglioniBDe MomiEFioriniPValdastriP. Autonomy in surgical robotics. Annu Rev Control Robot Autonomous Syst (2024) 4:651–79. doi: 10.1146/annurev-control-062420-090543

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 104

  HaideggerT. Autonomy for surgical robots: concepts and paradigms. IEEE Trans Med Robot Bionics (2019) 1(2):65–76. doi: 10.1109/TMRB.2019.2913282

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 105

  RobSurgical. Available online at: https://www.robsurgical.com/.

  • Google Scholar

• 106

  NguJC-YLinCC-WSiaCJ-YTeoN-Z. A narrative review of the Medtronic Hugo RAS and technical comparison with the Intuitive da Vinci robotic surgical system. J Robot Surg (2024) 18:99. doi: 10.1007/s11701-024-01838-5

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 107

  HeemeyerFBoehlerQKimMBendokBRTurcotteELBatjetHHet al. Telesurgery and the importance of context. Sci Robot (2025) 10(99):eadq0192. doi: 10.1126/scirobotics.adq0192

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 108

  Horizon Surgical Systems. Horizon surgical systems completes world’s first robotic-assisted cataract surgery [online] (2025). Available at: https://horizonsurgicalsystems.com/horizon-surgical-systems-completes-worlds-first-robotic-assisted-cataract-surgery/

  • Google Scholar

• 109

  ZhangPGehlbachPTaylorRHIordachitaJKobilarovM. Deep learning–based autonomous retinal vein cannulation in ex vivo porcine eyes. Sci Robot (2025) 10(109):eadw2969. doi: 10.1126/scirobotics.adw2969

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 110

  BianGBDengYLiZYeQZhaiYHuangYet al. Autonomous robotic intraocular surgery for targeted retinal injections. Sci Robot (2026) 11(110):eadx7359:1–17. doi: 10.1126/scirobotics.adx7359

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 111

  MenichiniGMalzoneGTamburelloSAndreoliALMoriFBallestínAet al. Safety and efficacy of Symani robotic-assisted microsurgery: assessment of vascular anastomosis patency, thrombus, and stenosis in a randomized preclinical study. J Plastic Reconstruct Aesthetic Surg (2024) 96:1–10. doi: 10.1016/j.bjps.2024.06.009

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 112

  DebeijGJonisYKaravoliaEHulstRQiuSSMulkenT. Robotic-assisted microsurgery using the MUSA-robot: evaluation of the learning curve in three clinical pilot trials. Plast Aesthet Res (2024) 11:48. doi: 10.20517/2347-9264.2024.51

  • CrossRef
  • Google Scholar

• 113

  MiHMacLarenRECehajic-KapetanovicJ. Robotising vitreoretinal surgeries. Eye (2025) 39:673–82. doi: 10.1038/s41433-024-03149-3

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 114

  KimHSOhB-YCheongCParkMHChungSSLeeRAet al. Single-incision robotic colorectal surgery with the da Vinci SP^® surgical system: initial results of 50 cases. Techniques Coloproctol (2023) 27:589–99. doi: 10.1007/s10151-023-02791-5

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 115

  MurguSDChenACConnellyJStermanDHYasufukuKLaxmananBet al. Safety results from multi-center observational real-world robotic bronchocopy (Target) study. Chest J (2023) 164(4):A5248–50. doi: 10.1016/j.chest.2023.07.3403

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 116

  AshrafianHClancyOGroverVDarziA. The evolution of robotic surgery: surgical and anaesthetic aspects. Br J Anaesthesia (2017) 119:72–84. doi: 10.1093/bja/aex383

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 117

  SridharANBriggsTPKellyJDNathanS. Training in robotic surgery—an overview. Curr Urol Rep (2017) 18:58. doi: 10.1007/s11934-017-0710-y

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 118

  ChenRRodrigues ArmijoPKrauseCSAGES Robotic Task ForceSiuK-COleynikovD. A comprehensive review of robotic surgery curriculum and training for residents, fellows, and postgraduate surgical education. Surg Endosc (2020) 34:361–7. doi: 10.1007/s00464-019-06775-1

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 119

  CofranLCohenTAlfredMKanjiFChoiESavageSet al. Barriers to safety and efficiency in robotic surgery docking. Surg Endosc (2022) 36:206–15. doi: 10.1007/s00464-020-08258-0

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 120

  PayneCJYangGZ. Hand-held medical robots. Ann BioMed Eng (2014) 42:1594–605. doi: 10.1007/s10439-014-1042-4

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 121

  ZhangTGongLWangSZuoS. Hand-held instrument with integrated parallel mechanism for active tremor compensation during microsurgery. Ann Biomed Eng. (2020) 48:413–25. doi: 10.1007/s10439-019-02358-2

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 122

  OstranderBTMassillonDMellerLChiuZ-YYipMOriscoRK. The current state of autonomous suturing: a systematic review. Surg Endosc (2024) 38:2383–97. doi: 10.1007/s00464-024-10788-w

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 123

  WagnerMBihlmaierAKenngottHGMietkowskiPScheiklPMBondenstedtSet al. A learning robot for cognitive camera control in minimally invasive surgery. Surg Endosc (2021) 35:5365–74. doi: 10.1007/s00464-021-08509-8

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 124

  El RassiIEl RassiJM. A review of haptic feedback in tele-operated robotic surgery. J Med Eng Technol (2020) 44(5):247–54. doi: 10.1080/03091902.2020.1772391

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 125

  BaltussenEJMBrouwer de KoningSGSandersJAalbersAGJKokNFMBeetsGLet al. Tissue diagnosis during colorectal cancer surgery using optical sensing: an in vivo study. J Transl Med (2019) 17:333. doi: 10.1186/s12967-019-2083-0

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 126

  NavarroSEMuhlbacher-KarrerSAlagiHZanglHKoyomaKHeinB. Proximity perception in human-centered robotics: a survey on sensing systems and applications. IEEE Trans Robot (2022) 38(3):1599–620. doi: 10.1109/TRO.2021.3111786

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 127

  TaketomiTUchiyamaHIkedaS. Visual algorithms: a survey from 2010 to 2016. IPSJ T Comput Vis Appl (2017) 9:16. doi: 10.1186/s41074-017-0027-2

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 128

  GuptaTGongWMaCPawlowskiNHilmkilAScetbonMet al. The essential role of causality in foundation world models for embodied AI. arXiv [preprint] (2024). doi: 10.48550/arXiv.2402.06665

  • CrossRef
  • Google Scholar

• 129

  VaccaroMAlmaatouqAMaloneT. When combinations of humans and AI are useful: a systematic review and meta-analysis. Nat Hum Behav (2024) 8:2293–303. doi: 10.1038/s41562-024-02024-1

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 130

  YanL. From passive tool to socio-cognitive teammate: a conceptual framework for agentic AI in human-AI collaborative learning. arXiv [preprint] (2025). doi: 10.48550/arXiv.2508.14825

  • CrossRef
  • Google Scholar

• 131

  Royal Academy of Engineering. People’s AI Stewardship Summit: scribe capture document (2025). Available at: https://raeng.org.uk/media/gg1nik4v/london-ai-summit-peoples-capture-document.pdf

  • Google Scholar

• 132

  GraceKSalvatierJDafoeAZhangBEvansO. Viewpoint: when will AI exceed human performance? Evidence from AI experts. J Artif Intell Res (2018) 62:729–54 doi: 10.1613/jair.1.11222

  • CrossRef
  • Google Scholar

• 133

  OffiahG. The liability challenges of AI. Bull R Coll Surgeons Engl (2025) 107(3):132–4. doi: 10.1308/rcsbull.2025.54

  • CrossRef
  • Google Scholar

• 134

  KhannaRRaisonNGranadosAOurselinSMontorsiFBrigantiAet al. At the cutting edge: the potential of autonomous surgery and challenges faced. BMJ Surg Interv Health Technol (2025) 7(1):e000338. doi: 10.1136/bmjsit-2024-000338

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 135

  KneeboneRL. Performing surgery: commonalities with performers outside medicine. Front Psychol (2016) 7:1233. doi: 10.3389/fpsyg.2016.01233

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 136

  AbramoffMDTarverMELoyo-BerriosNTrujilloSCharDObermeyerZet al. Considerations for addressing bias in artificial intelligence for health equity. Nat Digit Med (2023) 6:170. doi: 10.1038/s41746-023-00913-9

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 137

  NororiNHuQAellenFMFaraciFDTzovaraA. Addressing bias in big data and AI for health care: a call for open science. Patterns (2021) 2(10):100347. doi: 10.1016/j.patter.2021.100347

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 138

  HasanzadehFJosephsonCBWatersGAdedinsewoDAziziZWhiteJA. Bias recognition and mitigation strategies in artificial intelligence healthcare applications. NPJ Digit Med (2025) 8(154). doi: 10.1038/s41746-025-01503-7

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 139

  ZhangFKreuterDChenYDittmerSTullSShadbahrTet al. Recent methodological advances in federated learning for healthcare. Patterns (2024) 5(6):101006. doi: 10.1016/j.patter.2024.101006

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 140

  RehmanMHULopez PinayaWHNachevPTeoJTOurselinSCardosoMJ. Federated learning for medical imaging radiology. Br J Radiol (2023) 96(1150):20220890. doi: 10.1259/bjr.20220890

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 141

  YangJDungNTThachPNPhongNTPhuVDPhuKDet al. Generalizability assessment of AI models across hospitals in a low-middle and high income country. Nat Commun (2024) 15:8270. doi: 10.1038/s41467-024-52618-6

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 142

  ParkHJ. Patient perspectives on informed consent for medical AI: a web-based experiment. Digit Health (2024) 10:1–16. doi: 10.1177/20552076241247938

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 143

  World Economic Forum. Data equity: foundational concepts for generative AI. Geneva: WEF (2023). Available at: https://www3.weforum.org/docs/WEF_Data_Equity_Concepts_Generative_AI_2023.pdf

  • Google Scholar

• 144

  Responsible AI UK. Responsible AI. To enable flourishing in and by low- and middle-income countries (LMICs). Southampton: University of Southampton (2025). doi: 10.5258/RAi/001

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 145

  BalakrishnanGKrishnamurthyA. Robotic surgeries for cancer in low- and middle-income countries: hope or hype in surgical oncology? Int J Adv Med Health Res (2023) 10(2):64–70. doi: 10.4103/ijamr.ijamr_245_23

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 146

  TranVTRavaudP. Frugal innovation in medicine for low resource settings. BMC (2016) 14:102. doi: 10.1186/s12916-016-0651-1

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 147

  AlamiHRivardLLehouxPHoffmanSJCadedduSBMSavoldelliMet al. Artificial intelligence in health care: laying the foundation for responsible, sustainable, and inclusive innovation in low- and middle-income countries. Global Health (2020) 16:52. doi: 10.1186/s12992-020-00584-1

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 148

  Nature addresses helicopter research and ethics dumping. Nature (2022) 606:7. doi: 10.1038/d41586-022-01423-6

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 149

  OngJCLChangSY-HWilliamWButteAJShahNHChewLSTet al. Ethical and regulatory challenges of large language models in medicine. Lancet Digit Health (2024) 6(6):428–32. doi: 10.1016/S2589-7500(24)00061-X

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 150

  LeeABakerTSBedersonJBRapoportBI. Levels of autonomy in FDA-cleared surgical robots: a systematic review. NPJ Digit Med (2024) 7(1):103. doi: 10.1038/s41746-024-01102-y

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 151

  LefkovichCRothenbergS. Identification of predicate creep under the 510(k) process: a case study of a robotic surgical device. PloS One (2023) 18(3):e0283442. doi: 10.1371/journal.pone.0283442

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 152

  LeeBKramerPSandriSChandaRFavoritoCNasefOet al. Early recalls and clinical validation gaps in artificial intelligence-enabled medical devices. JAMA Health Forum. (2025) 6(8):e253172. doi: 10.1001/jamahealthforum.2025.3172

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 153

  HwangTJDasguptaP. Urgently clarify how AI can be used in medicine under new EU law. Nature (2024) 632:985. doi: 10.1038/d41586-024-02753-3

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 154

  MarcusHJRamirezPTKhanDZHorsfallHLHanrahanJGWilliamsSCet al. The IDEAL framework for surgical robotics: development, comparative evaluation and long-term monitoring. Nat Med (2024) 30:61–75. doi: 10.1038/s41591-023-02732-7

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 155

  LiuXCruz RiveraSMoherDCalvertMJDennistonAKThe SPIRIT-AI and CONSORT-AI Working Group. Reporting guidelines for clinical trial reports for interventions involving artificial intelligence: the CONSORT-AI extension. Nat Med (2020) 26:1364–74. doi: 10.1038/s41591-020-1034-x

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 156

  RiveraSCLiuXChanAWDennistonAKCalvertMJSPIRIT-AI and CONSORT-AI Working Group. Guidelines for clinical trial protocols for interventions involving artificial intelligence: the SPIRIT-AI extension. BMJ (2020) 370:m3210. doi: 10.1136/bmj.m3210

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 157

  ChenHGomezCHuangCMUnberathM. Explainable medical imaging AI needs human-centered design: guidelines and evidence from a systematic review. NPJ Digit Med (2022) 5:156. doi: 10.1038/s41746-022-00699-2

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 158

  Bracho-SanchezE. FDA warns about robotic devices for mastectomy and other cancer surgeries. CNN Health (2019). Available at: https://edition.cnn.com/2019/03/01/health/fda-warning-robotic-surgery-mastectomy/index.html

  • Google Scholar

• 159

  ZhangBZhangLChenQJinZLiuSZhangS. Harnessing artificial intelligence to improve clinical trial design. Commun Med (2023) 3:191. doi: 10.1038/s43856-023-00425-3

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 160

  AnuyahSSinghMKNyavorH. Advancing clinical trial outcomes using deep learning and predictive modelling: bridging precision medicine and patient-centered care. World J Adv Res Rev (2024) 24(3):1–25. doi: 10.30574/wjarr.2024.24.3.3671

  • CrossRef
  • Google Scholar

• 161

  WangZGaoCGlassLMSunJ. Artificial intelligence for in silico clinical trials: a review. arXiv [preprint] (2022). doi: 10.48550/arXiv.2209.09023

  • CrossRef
  • Google Scholar

• 162

  LiauMYQTohEQMuhamedSSelvakumarSVShelatVG. Can propensity score matching replace randomized controlled trials? World J Methodol (2024) 14(1):90590. doi: 10.5662/wjm.v14.i1.90590

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 163

  StrayhornJM. Virtual controls as an alternative to randomized controlled trials for assessing efficacy of interventions. BMC Med Res Methodol (2021) 21:3. doi: 10.1186/s12874-020-01191-9

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 164

  SinghRPaxtonMAuclairJ. Regulating the AI-enabled ecosystem for human therapeutics. Commun Med (2025) 5(1):181. doi: 10.1038/s43856-025-00910-x

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 165

  Medicines & Healthcare Products Regulatory Agency. Software and artificial intelligence (AI) as a medical device [online] (2023). Available at: https://www.gov.uk/government/publications/software-and-artificial-intelligence-ai-as-a-medical-device

  • Google Scholar

• 166

  EjazHMcGrathHWongBLGuiseAVercauterenTShapeyJ. Artificial intelligence and medical education: a global mixed-methods study of medical students’ perspectives. Digit Health (2022) 8:20552076221089099. doi: 10.1177/20552076221089099

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 167

  SethPHueppchenNMillerSDRudziczFDingJParakhKet al. Data science as a core competency in undergraduate medical education in the age of artificial intelligence in health care. JMIR Med Educ (2023) 9:e46344. doi: 10.2196/46344

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 168

  CarJOngQCErlikh FoxTLeightleyDKempSJŠvabIet al. The digital health competencies in medical education framework: an international consensus statement based on a delphi study. JAMA Netw Open (2025) 8(1):e2453131. doi: 10.1001/jamanetworkopen.2024.53131

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 169

  BabicBGlenn CohenISternADLiYOuelletM. A general framework for governing marketed AI/ML medical devices. NPJ Digit Med (2025) 8(1):328. doi: 10.1038/s41746-025-01717-9

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 170

  ChinzeiK. Safety of surgical robots and IEC 80601-2-77: the first international standard for surgical robots. Acta Polytechnica Hungarica (2019) 16(8):171–84. doi: 10.12700/APH.16.8.2019.8.10

  • CrossRef
  • Google Scholar

• 171

  IEEE Robotics and Automation Society. IEEE industry activities: approved standards [online]. Available at: https://www.ieee-ras.org/industry-government/standards/existing-projects

  • Google Scholar

• 172

  FengJEmersonSSimonN. Approval policies for modifications to machine learning-based software as a medical device: a study of bio-creep. Biometrics (2021) 77(1):31–44. doi: 10.1111/biom.13379

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 173

  VokingerKNFeuerriegelSKesselheimAS. Continual learning in medical devices: FDA’s action plan and beyond. Lancet Digit Health (2021) 3(6):e337–8. doi: 10.1016/S2589-7500(21)00076-5

  • Pubmed Abstract
  • CrossRef
  • Google Scholar

• 174

  BrojkaAQianAKrishnaKNurmalitasariNMadduriVYangY. Governance of adaptive artificial intelligence in medical devices. University College London (2025). Available at: https://www.ucl.ac.uk/engineering/sites/engineering/files/8_final_report_-_governance_of_adaptive_ai_systems.pdf

  • Google Scholar