The Challenge: Scaling Surgical Expertise

When I took on the role of Program Director for the Wellcome Leap SAVE program, I was aiming to solve some major challenges in surgery. Caring for surgical patients requires extensive training, careful decision-making, technical competence, and systems of care that can simultaneously manage resources and patient recovery. In collaboration with 16 teams comprised of over 300 researchers across 19 countries, we collectively worked to establish objective measures of competence, develop novel skills training platforms, and reduce postoperative mortality through advanced monitoring techniques. In addition, in just two years the SAVE network captured over two thousand surgical videos from over twenty facilities spread across eight countries on four continents. 

A critical barrier to unlocking surgery’s potential at scale is the need to train more surgeons in a way that ensures rapid skills development while also benchmarking and assuring technical proficiency. Yet we have not been able to train surgeons fast enough to keep up with population growth and medical needs: training has mostly flatlined while populations have surged. Furthermore, the volume of knowledge required and information available exceeds our human capacity. At this vulnerable moment, decision support systems can offer an enormous benefit to both clinicians and patients.

Bringing Aimbient Into SAVE

This insight led me to seek a team that could deliver on the promise of systems-level understanding of operative care in real time, the capacity to grow and leverage novel technologies as they become available from a multitude of sources, and a focus on safety and quality of care. I brought Aimbient into SAVE because they had the right combination of technical capabilities, clinical insights, a functional platform, a commitment to patient safety, and a vision of how decision support systems could evolve to improve care.

Aimbient not only brought its existing platform to bear, they also developed novel systems of video capture and annotation that would allow further work and exploration in the computer vision space. Importantly, they were willing to work with a diverse, multinational, multidisciplinary team to develop the capabilities to move surgical science and computer vision capabilities to the next level. 

The Complexity—and Value—of Surgical Video

Surgical videos are extremely challenging – they are generated in complex clinical environments, during time-sensitive care, through equipment that varies from one operating room to another. Video feeds differ in resolution, frame rate, color, field of view, and reliability. Procedures are performed by different teams, using different techniques and equipment, in hospitals with distinct workflows and variable resources.

Capturing this variation is not a nuisance to be eliminated. It is precisely what makes such a dataset useful.

Building an AI-Ready Surgical Video Library

The future of scalable safety in surgical care depends on technology that can learn from the full complexity of the operating room. Aimbient helped build that foundation by enabling SAVE to capture, manage, and store diverse surgical video across hospitals and countries for ORena, a surgical vision-language model dataset designed to advance the next generation of surgical AI.

While this may sound purely operational, it was central to the advancement of the science. Without reliable, ethical, high-quality systems for collecting real-world surgical data, there can be no genuinely diverse, AI-ready dataset – and no credible pathway for developing surgical AI tools that work across diverse institutions.

Aimbient put in place the resources needed to acquire and preserve surgical video from multiple hospitals and countries. This included a bespoke, secure platform for uploading surgical videos, secure server space for hosting sensitive videos, a mechanism for deidentification, an enhanced annotation platform, automated metrics to assess technical performance, and the ability to share these resources across multiple hospitals and settings. And all of this was done in ways that did not distract from clinical practice routines.

The Journey Forward, Launching the ORena FOCUS Challenge

Because I wanted SAVE to continue to inspire this diverse community of clinicians, computer scientists, researchers, and engineers, we launched the ORena FOCUS challenge, a vision language model challenge run by an expert team at the German Cancer Research Center which seeks to address difficulties in both contextual understanding of complicated surgical videos and the “remembering” of events over hours of time in a dynamic environment. 

But a challenge is only as valuable as the data and evaluation framework beneath it. If the underlying cases are too uniform, too clean, or too concentrated in a few settings, high model performance may say more about a model’s familiarity with and training on the dataset than its generalizability.

Designing Diversity Into the Data

Data diversity must be designed into the data pipeline from the beginning; it must be a feature of the dataset itself. This includes geography, hospital type, patient populations, procedure types, clinical teams, equipment, operative approaches, and operative techniques. It also includes the ordinary variability of real care: blood, smoke, camera occlusion, unexpected anatomy, and changes in camera position. While such features make the dataset harder to curate – and models harder to train – they are nevertheless essential if we want robust AI systems to function in real operating rooms.

A Lasting Legacy for Surgical Science

In just two years, SAVE created a deidentified surgical video library spanning more than twenty facilities across eight countries and four continents, enabling clinical and research teams to review, annotate, and share videos that can advance surgical science for years to come. Made possible by Aimbient’s capabilities and infrastructure, this work represents more than a global video library—it is a foundation for the next generation of surgical innovation and safer care for patients everywhere.