Artificial intelligence (AI) has the potential to redefine the surgeon’s role if medical professionals and technologists collaborate effectively.
The profession can prioritise data integrity and patient safety throughout this innovative process by maintaining realistic expectations about current capabilities and future advancements.
These key insights were highlighted in the recent webinar AI in Surgery, the fourth instalment in Surgery International’s Talk Surgery series.
Healthcare futurists and full-time surgeons offered an unmissable educational experience, highlighting how AI-based surgical systems can map an approach to each patient’s surgical needs and guide and streamline surgical procedures.
Hosted by Professor Shafi Ahmed, medical director of Surgery International, guests included:
• John Nosta, a leading innovation theorist in technology, AI and medicine.
• Sandip Sarkar, consultant vascular surgeon at Bart’s Health, who is conducting research using natural language processing to predict risk in diabetic patients.
• Thomas Knox, the founder and CEO of VitVio, which brings computer vision to the surgical operating theatre, creating solutions that improve patient outcomes and surgical efficiency. VitVio is the first platform built around AI and computer vision to make operating rooms safer and more efficient.
• Sirko Pelz, CEO of apoQlar, the developer of a medical mixed reality platform that is revolutionising how medicine is practised, experienced, learned and shared. VSI HoloMedicine is a medically certified software platform that leverages the Microsoft HoloLens hardware to transform medical images, clinical workflows and medical education into an interactive 3D mixed reality environment.
The global interest in large language models is immense, and surgery is no exception. Opening the discussion, Shafi Ahmed welcomed the webinar’s panel, noting: ‘We’ve gone from being an analogue surgeon to a digital surgeon in a matter of years, and the future looks very interesting and exciting.’
Sandip Sarkar is a consultant vascular surgeon at Bart’s Health. He used a natural language processing technique to code the whole of his diabetic surgery cohort to work out the simple premise of who is going to get worse with their diabetic foot disease and who he needs to concentrate his efforts on.
‘I was standing on the shoulders of giants and was very fortunate to get them to do this work for me. I stumbled into clinically using this in a triaging application, which has so many people interested in this AI project that I can use it clinically to find the patients we’re most worried about.’
Shafi suggested this is ground-breaking in using AI for surgical diagnosis and diagnostics and commented on how the world is one of convergence and collaboration with industry partners coming together with surgeons to build the future of surgery.
Sandip agreed that fundamental science is needed to ensure AI works appropriately.
He explained: ‘It’s still in its infancy, even though perhaps research was the first real element in medicine that fully used AI. So, the more mature elements of AI use in research are for trial design, recruitment and selection of patients, or selection of potential patients for clinical trials, and coding of data and results for research. And now, more and more, we’re moving into AI uses to look at all the research outputs we can now pull out, even without understanding the research outputs.
‘The very basics of research involve using good, reliable data, and using AI tools to ensure clean, reliable, standardised data is the other element, which, of course, is the same as what we want in clinical medicine.’
He discussed the complexities of treating patients with multiple medical conditions, particularly those who are diabetic with multisystem involvement.
Highlighting the challenge of reaching consensus in multidisciplinary team meetings where specialists, like nephrologists and dermatologists, may have conflicting treatment recommendations, he proposed using natural language processing (NLP) to analyse electronic health records (EHRs) comprehensively.
This method, he explained, has allowed them to identify diabetic patients and stratify those at risk of diabetic foot disease, leading to early interventions that can improve patient outcomes, such as increased longevity and reduced limb amputations.
The ultimate goal is to perfect the coding in EHRs to accurately predict patient pathways and tailor interventions effectively.
Thomas Knox talked about how his company is addressing the significant issue of overworked staff and backlogs in healthcare, particularly in the UK, by deploying cameras and sensors in surgical operating rooms to track people, equipment and processes.
He explained that the data is used to understand the context and stage of surgeries, which helps trigger timely actions around the hospital, such as prepping the next patient as a surgery nears completion.
This then ensures staff focus on patient care rather than data logging. He acknowledged the challenge of this transition but noted a shift in attitude post-COVID, with hospitals more open to innovation.
The initiative also aims to improve staff retention by removing undesirable tasks like data logging and avoiding adding steps to the surgical process.
He added: ‘We’ve heard some amazing stories… and if you were to implement something like this, you can help retain staff and keep them happier.’
Sirko Pelz discussed the integration of AI and holograms in medical imaging and procedures. They automatically upload DICOM files to create holograms from CT, MRI and PET scans. AI starts by anonymising patient data, especially in cloud environments. For surgical planning and segmentation, AI assists, but doctors must confirm results to meet certification requirements and avoid high-risk classifications.
He highlighted the progress in guided placement and navigation, where virtual markers can be set on holograms and patients. They have achieved near-instantaneous superimposition of images, like MRI scans, onto patients using AI for tasks like point cloud mapping.
He explained that the technology is advancing, aiming for automatic, real-time superimposition without external cameras. Future challenges include patient movement and organ tissue shifts during surgery, which AI will also manage. He predicts that in five years, such technologies will become standard in medical practice.
John Nosta discussed the evolution of surgery and medicine from an emphasis on manual technique to a focus on cognitive decision-making facilitated by advanced technology.
He predicted that AI and robotics will enhance surgical precision and decision-making, allowing more tasks to be delegated to technologically augmented technicians. This, he said, could lead to a hybrid model of surgical care, where AI augments human capabilities, potentially reshaping the traditional roles of surgeons and other medical practitioners.
He said:’ The fundamental expression of artificial intelligence and technology in general is artificially personified in the context of a robot – because they take on a human form, making it feel like a natural extension. I don't think you'll see the robot enter the operating room procedure, but we may see other humans have expanded skills. I think we're going to see AI augment the various levels of technical skills, creating a hybrid surgical intervener that will redefine the role of the physician.’
He also discussed the transition into the ‘cognitive age’, suggesting that large language models (LLMs) perform tasks faster and better, making processes enjoyable and stimulating cognitive engagement and creativity.
He also noted that LLMs help mitigate the monotonous aspects of clinical medicine, particularly those exacerbated by electronic health records, to foster a more engaging and intellectually stimulating professional environment.
Addressing the evolving approach to data in AI, the panel agreed on the need for clean, relevant data collection. Surgeons are advised to clearly define their AI objectives and gather only necessary, high-quality data, often starting with pilot studies using depersonalised datasets.
Integrating AI in clinical settings requires continuous validation and clinician oversight. A phased approach from non-clinical to clinical applications is recommended to build trust and ensure safety.
The speakers highlighted the importance of rigorous processes and regulatory compliance, acknowledging that while AI offers significant potential, it must be implemented cautiously to avoid undermining patient trust.
‘So, the only way to overcome that is by building confidence over time that it can potentially help stop human error,’ they concluded.
Finally, they acknowledged that ‘we have to take it step by step and don't expect things to be solved tomorrow’ and ‘be realistic about where we are now and where we might be in a few years’.
To watch the AI in Surgery webinar in full and for free, click here.
