In an operating room, the camera can record what the surgeon sees, a second camera can show what is happening around the table, and connected devices can record when instruments are used. Together, those signals can help a team understand why a case ran late or when the room will be ready for the next patient. They can also become training material for an AI model. The moment that model can be deployed at another hospital, a new question arises. Who can reuse what it learned in the first one?

On 16 November 2026, Richard Carter, CTO of Proximie, described how the company is adapting NVIDIA’s Cosmos 3 models using video, audio and device signals from operating rooms. He also describes a technique Proximie says it patented for obscuring identifying parts of video in the operating room before footage is sent elsewhere. The article spans operational uses such as room-readiness alerts and future concepts such as a second check before an implant is used. The two illustrate how different uses of the same infrastructure can place different demands on evidence and oversight. Proximie’s November publication on AI in surgery

Video by Proximie, published in 2022. The clip shows an earlier use of the platform in robotic surgery. Proximie in Action – Robotic Surgery Highlight.

Proximie’s collaboration with UC San Diego Health makes the deployment question tangible. The partners announced plans to roll out the platform in 44 operating and procedure rooms across four sites and to develop perioperative AI together. A model can learn from several rooms within one health system. Scaling that work to another health system brings different clinical routines, equipment and permissions into play. Five voices from the MedTech ecosystem help explain how those conditions shape the IP decisions. Proximie’s announcement of the UC San Diego Health collaboration

Nadine Hachach-Haram and the value of context from the operating room

When Proximie announced its collaboration with NVIDIA in April, founder and CEO Nadine Hachach-Haram emphasised the work involved in capturing and structuring surgical data. A video of the surgical field records part of a procedure. A wider view, together with instrument signals and timing information, can help explain which phase of the procedure the team has reached and what the room needs next. Proximie’s announcement of the NVIDIA collaboration

For a hospital and an AI developer, access for day-to-day operation and permission to train a reusable model are distinct decisions. A hospital may want the model to learn its local setup and scheduling routines. The developer may want to use some of the resulting improvements across customers. An agreement can make those uses explicit for footage, annotations and trained model components, including which uses may continue after the partnership ends.

Bjoern von Siemens and the work of assembling surgical data

In a published contribution on surgical AI, caresyntax co-founder Bjoern von Siemens points to a practical bottleneck. Surgical data are difficult to assemble in a form that teams and developers can use to improve the next case. Footage, instrument events and records of what happened during a procedure need to be connected before they can support useful analysis. Published contribution on surgical data and foundation models

That work creates assets of its own. A hospital may contribute the expertise used to label phases of a procedure. A device maker may explain what its logs record. A platform developer may build the method that synchronises signals from different sources. Which of those contributions can each partner reuse? The answer matters when the first deployment becomes the starting point for a product sold to other hospitals.

Rachel Dunscombe and the interfaces that let information travel

In an interview about preparing health data for AI, HL7 International CEO Rachel Dunscombe discusses making standards such as FHIR easier for AI systems to ingest. The operating-room example shows where standards can help: an alert or procedure record has to reach another system with enough context for the receiving team to understand it. Interview on HL7’s role in preparing data for AI

A shared format can make that connection easier to implement. A hospital still needs access to the relevant interface, documentation and historical data in a usable form. If a device is replaced or the hospital changes platform, those practical permissions can determine whether the team can keep using the workflow and carry its records forward.

Mark Slack and evidence that follows deployment

In a 2023 paper, Mark Slack and co-authors describe a prospective registry created to collect real-world evidence as the Versius surgical robotic system entered clinical use. The registry offers a parallel for the intelligent operating room. When a technology moves into new settings, its developers and users need a way to learn how it performs under those conditions. Published study on the Versius registry

For an AI system, a change in equipment, workflows or the model itself can affect which evidence is relevant. A hospital considering a second deployment will want to know which version was evaluated, where it was used and which results can be shared to support its decision. Access to that evidence can become as consequential for expansion as access to the model.

Camille Terfve and the parts of the system worth protecting

In our CEIPI IP Business Talk, Camille Terfve discussed choosing protection according to the contribution each asset makes to a digital MedTech business. The Proximie case makes that choice tangible. In-room de-identification, the method that aligns device signals with video and the annotations created with clinical teams each contribute something different. CEIPI IP Business Talk on IP strategies for Digital MedTech

A company may seek patent protection for a technical method, manage implementation know-how confidentially and agree separately on how jointly developed assets may be used. For both hospital and supplier, the commercial question is which of those assets they will need to deploy the system again and who can give them the necessary access.

Bringing the ecosystem back to the IP decision

Before a hospital and its partners turn a first deployment into a wider programme, they can ask:

  • What data and clinical know-how can each partner use to operate, evaluate or train the system?
  • Which annotations, model adaptations and workflow improvements can be reused at another hospital?
  • What must stay local, and what information can be shared for training or evaluation under the agreed conditions?
  • Which device interfaces and historical records must remain accessible if a supplier or platform changes?
  • What evidence does the next site need, and who can share it when the model or its intended use changes?
  • Which technical contributions should be protected through patents or confidentiality, and which permissions should be settled in the collaboration agreement?

The first deployment establishes how the system fits into a working operating room. The IP and collaboration terms determine how much of that experience can support the next one. This is part of the wider discussion in our MedTech Industry Focus on the IPBA Plattform.