“Soon, AI will be able to perform surgery independently.”
As a PGY-4 vascular surgery resident conducting AI research, I encounter this prediction somewhat frequently. I have to wonder — do we (surgeons) want this? Autonomous surgery could potentially expand access to surgery; patients in certain areas of the United States and worldwide often struggle to access timely, high quality surgical care. Still, on a personal level, I would hesitate to relinquish the defining aspect of my profession to AI.
For the past year, I have had the privilege of working in Stanford’s Center for Artificial Intelligence in Medical Imaging with talented computer scientists from every corner of the world. The lab has shown me how AI can provide unique patient benefit, from making tumor board recommendations personalized to a patient’s specific genetics and tumor distribution, to using imaging to predict which patients have a genetic connective tissue disorder (my project, stay tuned!). At the same time, AI coding tools have allowed tech companies to eliminate much of their workforce; software engineers who have been lucky enough to avoid layoffs have needed to make radical alterations to their skillsets to stay employed.
The employment impact of this transformative technology is well known (and feared) in tech circles, but the impact of AI will certainly spread to every profession, including vascular surgery.
As with any new technology, we face the choice to adapt or be left behind. However, I contend that a secret third option is best for us and our patients: We must actively shape AI to fit our needs.
AI solutions for healthcare are typically not built by doctors. And for good reason — physicians are usually busy treating patients. Although our lab has a variety of physician collaborators who offer valuable guidance on building clinically useful, safe models, many commercial healthcare-adjacent AI solutions seem to be built from the perspective of engineers who experience health care as patients. For instance, Doctronic, which performs fully autonomous medication refills (without a medical license), could plausibly have arisen from frustration with wait times for a PCP medication refill appointment.
On our current trajectory, non-surgeons who imagine AI for surgery think of an autonomous AI surgeon. I am optimistic about the benefits of AI, but I do not think that this is a good goal — medicine (surgery in particular) will always be an art. Humans are not machines and the practice of alleviating suffering requires the care and trust that a human physician should provide. However, I think there are many opportunities for making vascular surgery safer, faster and more efficient using AI.
Consider a model that takes in a patient’s CT and chart, simulates peripheral vascular interventions and shows predicted patency rates for each option. Or perhaps an AI companion that our patients bring home to help coordinate care appointments, dispense medications and track changes in symptoms. We could develop an interactive AI cath lab that can anticipate steps of a procedure, ensures the right tools are in the room at the right time and learns our procedural preferences. Our unique perspectives as practicing surgeons allow us to imagine solutions that improve care without eliminating ourselves.
Among medical specialties, vascular surgery is often at the forefront of innovation and several surgeon-led/surgeon-advised companies are already bringing AI into vascular practice. As the AI revolution continues to pick up steam, it will be more important than ever for us to guide developers and engineers to make the right solutions. Even those who are not involved in software/device development should strive to cultivate a working familiarity with AI to prepare for the inevitable changes coming for clinical practice. I am hopeful we can continue to make AI work for us and not the other way around.
Andrea Fisher is a PGY-5 vascular surgery resident at Stanford, currently completing two years of research in Stanford’s Center for Artificial Intelligence in Medicine & Imaging and obtaining a master’s degree in biomedical data science.
