In 2024, more than 2.3 million adults in the U.S. reported having had a low-dose chest CT scan to check for lung cancer.1 Now imagine a future in which doctors can leverage appropriately validated AI tools so that a scan performed for that one purpose could also help assess signs of emphysema, identify thyroid nodules, assess imaging markers associated with cardiovascular risk, or even flag reduced bone density in the spine for further evaluation. Some of that information may already be captured, depending on the scan protocol, anatomy covered and image quality. Making sense of it could help inform further evaluation and preventive care.
This is the future of healthcare that GE HealthCare’s new Chief AI Officer, Rodolphe Katra, wants to bring to life. It’s a mission that’s deeply personal for Katra, who lost both of his grandfathers to heart disease.
Until recently, Katra had devoted much of his career to bringing healthcare innovations—from surgical technologies to ear, nose and throat diagnostics—to market. However, his interest in personalized disease discovery and care led him to take on the role of Chief AI Officer at GE HealthCare.
“I believe that with our installed base and our strong legacy of AI innovation, GE HealthCare is well positioned to help make the vision of precision care a reality.”
Based on the FDA’s 2026 list of AI-authorized medical devices2, GE HealthCare had the largest number of listed authorizations. The company’s broader portfolio spans imaging devices, disease-specific applications and software designed to help health systems operate more efficiently.
For example, Sonic DL 3D uses deep learning to support sampling acceleration factors of up to 12 and acquisition-time reductions of up to 86% compared to fully sampled MR datasets, depending on the protocol.3 These are acquisition-level measures, not reductions in the duration of an entire MR examination.
On the operational front, CareIntellect for Operations provides forecasts of potential bottlenecks up to 72 hours in advance, with recommendations for teams to review as they plan for discharges, imaging, transfers and other capacity needs.
In this conversation, Katra talks about precision care and his vision for the future of AI in healthcare, while also addressing the larger societal concerns about AI safety.
Precision care gets talked about a lot. What does it mean to you, and in your new role for GE HealthCare, where do you see the most opportunity?
Precision care starts with treating the person, not simply the disease. Today, I believe we are closer than ever to realizing that vision of personalized care, with GE HealthCare working toward that goal.
I’ve been tremendously impressed with how GE HealthCare has used AI to support clinicians in their work. Take Vscan Air with Caption AI, which is designed to guide appropriately trained healthcare professionals in maneuvering the probe to capture diagnostic-quality standard ultrasound views of the heart. I grew up in a healthcare system that was highly fragmented and relied on scarce clinical specialists to perform required diagnostic testing. My family would have loved that kind of guidance in an area where specialists were so hard to find. My hope is that guidance like this can help more appropriately trained clinicians acquire these images closer to where patients live.
When it comes to precision care, we can talk about both the short and the long term.
In the short term, I’m excited about connecting these tools to help clinicians care for patients across their health journey. As a patient moves from radiology to hematology and receives nursing care along the way, pieces of their medical story remain in separate systems. Bringing that information together, with appropriate permissions and safeguards, could help care teams make decisions with more of the patient’s history in view.
Over the longer term, I’m especially interested in the potential of opportunistic screening: using an exam performed for one reason to look for subtle signs of other conditions. Imagine you have a chest CT for lung cancer screening or to investigate a persistent cough and possible infection. With appropriately validated tools, AI could help clinicians assess additional findings in that same scan before symptoms appear, including calcium deposits on a heart valve or an enlarged heart that merit further evaluation. A cardiologist could then evaluate the findings and discuss appropriate prevention or treatment options with the patient. The value of this approach would need to be weighed against false positives, missed findings and unnecessary follow-up.
Of course, this kind of future will require a lot more than technological innovation. Getting there will require all the key players—companies with deep domain expertise like GE HealthCare, healthcare professionals, academics, technology companies, government regulators and payers—to work together and evolve the way we operate on a day-to-day basis.
The good news is that we’ve already integrated other forms of prevention into routine care. Today, recommended childhood vaccination helps protect children against certain infectious diseases. We have made that preventive measure part of routine healthcare and established ways to pay for it. AI-enabled screening would need its own supporting evidence and applicable coverage and payment decisions.
We’ve done this before. And I want GE HealthCare to help lead the work of making evidence-based preventive care more accessible.
“I’m especially interested in the potential of opportunistic screening: using an exam performed for one reason to look for subtle signs of other conditions.”
This is a space that moves fast: how do you see AI in healthcare evolving over the next year or so?
Over the near term, I think we’ll see progress in three areas.
The first is helping us personalize care with the goal of improving patient outcomes. Historically, we’ve focused on hard measures and endpoints: things like death, hospitalization and other objective outcome measures. Think of the typical measures you track in a clinical trial. AI’s ability to identify patterns in qualitative and numerical data could help us consider a broader range of factors alongside those clinical endpoints.
Take spine surgery. We’ve traditionally focused on correcting the spine according to established medical standards. But it’s also worth asking the patient: What do you want to be able to do after surgery? If someone wants to get back to running marathons, that ambition could help shape the surgical plan. Someone who uses a wheelchair might say, “What I really want is to get through the day without pain.” Those conversations are already part of patient-centered care. The opportunity is to explore how AI might help clinicians consider those goals alongside clinical evidence—not replace their judgment or the patient’s preferences.
The second area is helping clinicians deal with cognitive overload. Healthcare generates substantial amounts of digital data. Yet much of that information can be difficult to bring together and analyze across systems. When I speak to physicians, they tell me they need help making sense of all of this data. They want some kind of summary or insights layer to be able to see what’s relevant to the patient in front of them without having to sift through every record themselves, while retaining access to the underlying records when needed.
The third area is working toward broader healthcare access and more efficient care delivery. Today, health systems are facing rising demand, staffing shortages and increasing costs. That adds to the challenge of caring for more patients with available resources. Here’s where AI-powered offerings like CareIntellect for Operations are intended to support capacity planning and resource coordination. Whether that translates into greater access or lower costs depends on implementation and local workflows.
If we are to realize the potential of AI in healthcare, we have to address all of these needs together. We focus on the patients, the providers, and also on working to make precision care more affordable and available beyond those who can afford to pay more.

We’re living through a time where AI in healthcare shows a lot of promise—however, at the same time, there are a lot of concerns about AI and safety. How should healthcare leaders think about implementing AI within their organizations?
I get asked this question all the time, and my answer is always the same: patient safety isn’t something we can trade away to move faster with AI.
For healthcare leaders, that means evaluating evidence for the intended use and patient population, understanding limitations, training users, maintaining appropriate human oversight and monitoring performance after deployment. Access to patient information and its use also need appropriate privacy and governance safeguards.
I also want to stress that AI is just a means to an end. At its core, it’s math. It may be exciting math, but we need to judge it by whether it helps us achieve our goals: better patient outcomes, less cognitive overload for clinicians and a healthcare system that’s more affordable and accessible.
At GE HealthCare, we’ve built our legacy of trust over more than 125 years. One thing I’ve learned since joining is how seriously people here take that responsibility. Our Responsible AI principles guide how we develop products. Depending on the application, approaches such as visual grounding can help link an AI-generated response to relevant regions in an image, while ontology-based reasoning draws on structured knowledge about medical concepts and how they relate to one another. Our principles aren’t abstract sentiments—on the contrary, they help inform our approach to product development, evaluation and oversight.
- https://pmc.ncbi.nlm.nih.gov/articles/PMC13084730/ ↩︎
- https://www.gehealthcare.com/en/about/newsroom/press-releases/ge-healthcare-tops-list-with-highest-number-of-ai-enabled-medical-device-authorizations?srsltid=AYCX9vL0JjgUtG91zmw4HACm88wxN6WpkWsmhFN4Q3FwGUfXuLZZU-Ek ↩︎
- https://signapulse.gehealthcare.com/autumn-2024/introducing-sonic-dl-for-highly-accelerated-isotropic-acquisitions ↩︎



