CARE™, Aidoc’s Clinical AI Reasoning Engine, is a foundation model trained on rich, real-world multimodal data. It not only accelerates the development of new use cases but also enhances the performance of existing FDA-cleared applications — making them smarter, faster and more adaptive.*
To explore what this means for healthcare, we sat down with our Chief Product Officer, Reut Yalon, PhD, to discuss the origins of CARE™, how it differs from Aidoc’s legacy algorithms and its impact on the future of patient care.
RY: It’s a fair question, but I want to be clear this isn’t about us suddenly re-evaluating our approach. What really happened was, until recently, we were developing algorithms with deep learning technology, before the foundation model era.
For every disease state, whether it was for triage, classification or measurement, we built a separate algorithm. That meant collecting variable data, annotating it carefully, training the model, validating it, building the product, submitting it to the FDA and only then commercializing. This process took one to two years for a single solution, and we had to repeat it for each and every algorithm.
Even with multiple R&D squads working in parallel, we always asked ourselves: Is this scalable enough? The truth is, it wasn’t. If you want to close preventable care gaps across the health system — when over 350,000 deaths each year are tied to diagnostic errors — doing it one algorithm doesn’t solve the problems.
Our dream was to create a model that could identify many different findings and perform many different tasks at once, but the technology wasn’t there. If you’d asked me a few years ago to build such a model, I would’ve said it was impossible.
That changed with foundation models. With CARE™, we can build, for each modality and anatomy, a generic model that can then be applied and tailored for numerous disease states or tasks, enabling us to develop a wide range of applications much faster and potentially with better accuracy.
Q: What makes CARE™ more intuitive than traditional deep learning models?
RY: CARE™ is more intuitive because it mirrors how clinicians think. Radiologists don’t just look at one pixel or one finding — they interpret the whole study, compare priors, review history and then characterize findings. CARE™ takes that holistic approach, too. It’s not a niche model that does one thing; it’s designed to manage multiple findings in context, whether they’re absent or present.
Q: How are CARE™-derived algorithms different from our legacy algorithms?
RY: The main difference is the underlying technology. With legacy AI, we had to train models on specific findings with pixel-level annotations. CARE™, as a foundation model, learns from very large datasets, and because it uses self-supervised learning, it doesn’t require every pixel to be annotated.
That shift allows the model to be more generalizable and development more scalable. It can detect, measure and characterize findings across many tasks, rather than just one. The data assets we’ve built over the years remain critical, but it’s the technology that unlocks this new approach.
Still, models dveloped on CARE™ – like any clinical algorithm – need to be delivered into actual real-world physician workflows, and that’s the role of Aidoc’s aiOS™, our enterprise-grade clinical AI platform.
Q: How do you respond to skepticism about foundation models being overhyped?
RY: With any major technological shift, there’s going to be hype. Some of what we see in the market won’t translate into real products. That’s normal.
What matters is discipline. At Aidoc, we don’t just build models, we validate rigorously, design workflows, ensure regulatory pathways and prove clinical value. We’ve already secured FDA clearance for CARE™-powered applications, and we’re actively working with regulators to define the path for foundation models, including breakthrough device designation.
While the excitement is real, so is our discipline. CARE™ isn’t about chasing buzzwords.
Q: What are the benefits to a health system?
RY: CARE™ enables health systems to expand AI coverage more broadly, improve accuracy and depth plus deliver results faster. Importantly, it also allows us to start partially automating pieces of the radiology workflow. That means not just flagging one or two findings but supporting radiologists across triage, detection, measurement, characterization and, finally, taking those elements to generate an AI-based preliminary report.
Q: Is CARE™ only for imaging?
RY: Imaging is where we started and where the first applications of foundation models make the most sense, but CARE™ isn’t limited to imaging. In the future, our models will be able to consume both imaging and non-imaging data — like EHR notes and labs — just like radiologists use multiple data points when interpreting studies today.
Q: How does CARE™ reflect the evolution of Aidoc’s product strategy?
RY: Our strategy has always been about reducing preventable care gaps with AI. Then we moved downstream into acute care coordination and patient management.
CARE™ strengthens this trajectory in two ways: It enables us to scale faster and more accurately across disease states, and it allows us to start building a fourth pillar — prediction. That’s the natural endpoint of our strategy: not just detecting and coordinating care but helping prevent disease altogether by spotting subtle patterns years before symptoms appear.
Q: What’s the clinical value of a foundation model like CARE™?
RY: There are three main benefits:
- Comprehensiveness – a broader suite of findings and tasks covered
- Depth – beyond detection, it can measure, characterize and compare findingm.s and generate a preliminary report
- Speed – use cases can be developed and deployed much faster
This means health systems see value sooner and across more areas of care. That’s what it will take to move from incremental gains to exponential impact on diagnostic quality.
Q: What excites you most about CARE™?
RY: Honestly, it’s the fact that our dream can finally become a reality. We’ve always wanted to meaningfully reduce clinical burden and to close care gaps.
Until now, we could only chip away at that vision one algorithm at a time. CARE™ makes it possible to do this comprehensively, more accurately and faster than ever before. And beyond radiology, it opens the door to embedding AI across the care continuum, so physicians can deliver the care they know their patients need.
Interested in learning more about CARE™?
Meet with our team.
*under new FDA clearances
