Foundation AI Models: The Next Revolution in Cardiovascular Medicine

Foundation AI models are transforming cardiovascular medicine with smarter diagnosis, precision treatment, and better heart outcomes.

Despite all the recent improvements in prevention and treatment, cardiovascular disease remains the leading cause of death in the world and a large proportion of the damage is done without the person ever knowing. A normal routine ECG can contain signs of heart failure, valve disease and dangerous muscle thickening for many years.

The FDA has approved an AI system called EchoNext on June 23, 2026, which can identify six of these hidden conditions from a simple 12-lead electrocardiogram (ECG), the same test that doctors would conduct in their offices on a daily basis. This one approval is just a sign of the times that cardiology is on the verge of entering the foundation model era.

It explains what foundation AI models are, how the science is already being applied in actual healthcare settings, and the future of the field without the hype or the jargon. 

What Is a Foundation AI Model?

A foundation AI model is a large-scale AI system that has been trained on a large, diverse data set to enable it to perform a variety of tasks. Imagine it was like a calculator versus a smartphone – one is designed for just one purpose while the other is a flexible platform you can customize for a variety of purposes.

For cardiovascular medicine, one example of a foundation model could be trained on millions of ECGs, echocardiograms, cardiac MRI, wearable sensor data and clinical notes, all simultaneously. After training, it can be fine-tuned for particular tasks such as identifying arrhythmia, assessing the risk of heart failure or identifying structural abnormalities that the human eye would not be able to recognise.

It is similar to the way that large language models such as ChatGPT are trained by analyzing vast amounts of text data to understand general language patterns and then fine-tuned for specific tasks. Cardiac foundation models function in the same way, but with the "language" being heart signals and images rather than words. 

Foundation Models vs. Traditional Cardiac AI

Training Scope

Traditional (Narrow) cardiac AI: This is the type of AI that is trained on a specific dataset and a specific task, e.g., identifying atrial fibrillation. Only trained on large and varied datasets for a variety of cardiovascular tasks – Foundation AI Models.

Data Types

Traditional (Narrow) Cardiac AI: Single-modality data (ecg, imaging, cardiac).

Foundation AI Models: Integrates multimodal data, such as ECGs, cardiac imaging, clinical notes, electronic health records, lab results, and data from wearable devices.

Adaptability

Traditional (Narrow) Cardiac AI: Must be retrained for a new task.

Foundation AI Models: Can be quickly fine-tuned for new clinical applications using relatively small additional datasets.

Generalization

Traditional (Narrow) Cardiac AI: These work well if the data it is working with is similar to what it was trained with but may not perform well on other hospitals or devices.

Foundation AI Models: Specially built to learn broadly from a variety of patients, healthcare systems and medical devices.

Example: Traditional (Narrow) Cardiac AI: An algorithm that was created to identify a particular arrhythmia, like atrial fibrillation.

Foundation AI Models: A model trained on millions of ECGs that can classify over 150 diagnostic categories for the heart and which can be used for various cardiovascular tasks. 

Inside the New Generation of Cardiovascular Foundation Models

ECG Foundation Models

The electrocardiogram is the oldest, most widely used test in cardiology and an ideal starting point for artificial intelligence. Created from more than 10 million ECGs from more than 1.8 million patients with real cardiologist annotations across approximately 150 diagnostic categories, ECGFounder, described in NEJM AI, was constructed. Individually, several researchers have introduced transformer-based ECG models for heart and coronary function assessment, based on ECG waveform, without requiring any supervision.

The clinical potential is great. An ECG is much cheaper than an MRI or echocardiogram, and can be recorded almost anywhere, including on some consumer-wearable devices. If AI can find even more insights in that cheap, widely available test, it will have the opportunity to greatly enhance access to sophisticated cardiac screening, especially in areas where it's not offered.

Multimodal and Imaging Foundation Models

The most ambitious are systems that integrate multiple data types simultaneously. The Cardiac Sensing Foundation Model (CSFM) is the February 2026 cover feature in Nature Machine Intelligence, led by researchers from the University of Oxford. CSFM was trained using cardiac data from approximately 1.7 million people, including physiological signals such as cardiac data taken from the smartwatch (ECG, PPG), clinical or AI-generated reports. When evaluated, it beat traditional "one-task, one-modality" models for all four tasks: diagnosis, demographic estimation, vital-sign prediction, and outcome forecasting.

In isolation, the AI has also been trained with contrastive learning, matching medical imaging scans with the corresponding text report that is written about them, as described in the Nature Biomedical Engineering paper. 

From Lab to Clinic: Real-World Applications Today

Heart Disease Risk Prediction

The use of foundation models to estimate the future cardiovascular risk of a patient, rather than just diagnose a current condition, is increasing. They are able to identify more subtle risk signatures years ahead of symptoms because they are based on patterns in wider populations.

Cardiac Imaging AI

An imaging platform recently cleared by FDA (in January 2026), AI-CVD can opportunistically extract cardiovascular risk measurements from routine chest or abdominal CT scans, which were typically ordered for other reasons, allowing tens of millions of annual scans to become a screening opportunity.

Clinical Decision Support in Cardiology

Most of the tools that have been cleared are meant to assist physicians, not to take the place of them. In the case of EchoNext, for example, it is being embedded in platforms such as OpenEvidence to enable AI-flagged findings to be displayed within the physician's workflow without the need for another platform. 

Regulatory Status: What's Actually FDA-Cleared Today

It's important to separate research breakthroughs from tools doctors can legally use on patients right now.

EchoNext

Status (as of mid-2026): FDA-cleared (June 23, 2026)

What it does: It applies AI to recognize six structural heart disease conditions from a standard 12-lead ECGs and avoids the need for an echocardiogram as a first-line screening tool.

AI-CVD (HeartLung Corp.)

Status (as of mid-2026): FDA-cleared (January 2026)

What it adds: Opportunistic screening of the heart with routine computed tomography of the chest that helps identify undiagnosed cardiovascular disease.

AliveCor's Kardia 12L / KAI 12L System

Current status (mid-2026): FDA approved, expanded indications (January 2026)

What it does: Helps interpret portable 12-lead ECGs with AI assistance beyond the normal cardiac determination of rhythm.

Ultreon 3.0 (Abbott)

Status (as of mid-2026): FDA-cleared (April 2026)

The purpose: Enhances cardiologists' ability to precisely size and position stents during percutaneous coronary intervention (PCI) using AI-powered coronary imaging.

CSFM, ECGFounder, and Cardiac MRI Foundation Models

Status (mid-2026): Peer-reviewed research; but not independently FDA-cleared as commercial products.

What they find: They show how well foundation AI models trained with large data sets of ECG and cardiac MRI – using AI to assist with a variety of cardiovascular use cases, such as disease detection, risk prediction and clinical decision support for multiple modalities. These models are the next generation of AI research in the field of cardiology. 

As of March 2026, cardiology ranked second only to radiology among all medical specialties for FDA-cleared AI algorithms, with well over 140 dedicated clearances.

Benefits for Patients, Clinicians, and Health Systems

Earlier detection: Overt CHD without apparent signs can now be picked-up from a test that many patients get on a regular basis.

Expansionary use: ECGs and in some instances, even wearables are low-cost and mobile, meaning that AI powered analysis could take advanced screening beyond clinics with on-site cardiology providers.

Efficiency: Pre-screening of patients can be automated to prioritize patients that are the most urgent for specialist referral or echocardiogram.

More complete risk pictures: Multimodal models using imaging signals and health records can provide a more comprehensive risk picture than any single test. 

Limitations, Risks, and Open Questions

The caveats cannot be ignored in the scrutiny of this area. It is important to point out that even at this stage, the use of foundation models is not a substitute for clinical judgment but rather a tool for decision support, based on evidence from reviews in the European Heart Journal – Digital Health and others. Other open questions are:

Bias and representation. The study, published in Frontiers in Artificial Intelligence, points to the need for caution when developing AI and precision-medicine technologies on datasets that lack diversity to ensure that racial and ethnic inequities in cardiovascular care are not reinforced in automated tools.

Cross-device and cross-population generalization. A model developed primarily from data from one hospital system might not work as well in another.

Lag in regulatory and reimbursement processes. With increasing clearances, reimbursement pathways for insurance coverage of AI-ECG analysis are still emerging.

Interpretability. Clinicians should be aware of the rationale behind a model's flagging, rather than just accept the result. 

The Future of Precision Cardiovascular Medicine

The direction of travel is increasingly described as moving from "cardiology as a disease diagnosis" to "cardiology as a health trajectory prediction and planning process," and sometimes in conjunction with the notion of a "digital twin" of the heart for personalized simulation and planning. The future looks bright and will rely on answers to today's open questions of bias, validation, and integration — and the rate of peer-reviewed publications and FDA activity during the first half of 2026 indicates the future is already the present, rather than merely imagined.

Frequently Asked Questions

1. What is meant by "foundation model" in AI, in simple terms?

A foundation model is a large-scale AI model that is trained on vast and diverse sets of data, and can be fine-tuned to perform diverse tasks. For cardiology, it would be one model that could be trained on ECGs, imaging, and clinical notes and then fine-tuned for a specific purpose, such as risk prediction or diagnosis.

2. What are some current uses of AI in the field of cardiology? 

AI is already being used to assist in the interpretation of ECGs, analysis of echocardiograms, coronary imaging, arrhythmia detection, and remote cardiac monitoring. In 2026, there were more than 140 FDA-cleared AI algorithms in cardiology than any other specialty (excluding radiology).

3. Is it possible for AI to predict a heart attack?

AI can detect patterns of risk and hidden structural defects that can predict cardiovascular events, potentially years in advance. It cannot predict a particular heart attack but on the whole can usefully help to stratify risk.

What is an ECG foundation model? A foundation model for ECG is an AI system trained with millions of electrocardiograms to identify a wide variety of cardiac patterns. For instance, ECGFounder has been trained on more than 10 million ECGs and approximately 150 diagnostic categories.

5) Does AI for cardiology have FDA approval?

Yes, under certain circumstances. AI-CVD and AliveCor's Kardia 12L, for example, have been FDA cleared for specific uses, as have tools such as EchoNext (expected for clearance in June 2026). A wider range of foundation model studies remains in publication and transition to commercial clearance.

6. What are potential pitfalls of heart disease diagnosis using AI?

Main risks involve the possibility of bias from non-diverse training data, less performance on different populations or devices, and over-reliance on the AI outputs without proper clinical interpretation and supervision.

7. How accurate is the accuracy of AI when compared to cardiologists reading ECGs? 

Few tools have been cleared, such as EchoNext, which has been reported to outperform cardiologists in certain validation studies, including with the help of AI. But there are different levels of accuracy across tools, tasks, and populations, so don't assume accuracy of results across all AI systems.

8. What is Multimodal AI in healthcare?

Multimodal AI brings together multiple data modalities, including imaging, waveform signals and text reports, in a single model. In cardiology, this enables a system to take into account an ECG, an echocardiogram, and clinical notes all at once – instead of analysing them separately.

9. Will Cardiologists Be Replaced By AI?

The latest evidence and regulatory approvals suggest that AI is more of a decision support tool, rather than a substitute for clinical judgment or the doctor-patient relationship.

10. Is there a place I can inquire if a clinic or hospital uses AI in its cardiac tools? 

You can specifically ask your care team if the ECG, imaging, or monitoring is AI-assisted, and you can review the public list of FDA cleared AI/ML-enabled medical devices on the FDA's website to see if there are any tools cleared for cardiovascular use. 

Conclusion

The Foundation AI models are general, multi-task systems, which are trained from both ECG data, imaging and clinical text, unlike older, single-task cardiac algorithms.

In 2026, real, peer-reviewed models will already be published in top journals: CSFM, ECGFounder, cardiac MRI foundation models. Currently, real FDA-cleared products are available to patients (EchoNext, AI-CVD, Kardia 12L, Ultreon 3.0).

Issues of bias, interpretability, and validation have yet to be resolved and are an issue in the responsible adoption. This transition is analogous to the leap from narrow, single-purpose tools to a more general-purpose system, and is also described in the same language that’s typically used to describe LLM systems — the "heart language".

Looking ahead, the FDA clearance market is projected to expand further, with a convergence with routine clinical practice, such as ECG carts and reference platforms, and a gradual transition from reactive diagnosis to predictive personalized cardiovascular care.

Call to Action: Clinicians and researchers: Check out the peer-reviewed models here on NEJM AI and Nature Machine Intelligence. For those who are a patient, ask your provider if AI-powered tools are included in your cardiac screening — and keep in mind as this technology continues to transition from the research lab into everyday healthcare.

References:

• FDA AI/ML-Enabled Medical Devices list: https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-and-machine-learning-aiml-enabled-medical-devices
• NEJM AI (ECGFounder): https://ai.nejm.org/doi/abs/10.1056/AIoa2401033
• Nature Machine Intelligence (CSFM, Feb 2026): https://www.nature.com/articles/s42256-026-01180-5
• Nature Biomedical Engineering (cardiac MRI foundation models):  https://www.nature.com/articles/s41551-026-01638-2
• European Heart Journal – Digital Health (review): https://academic.oup.com/ehjdh/advance-article/doi/10.1093/ehjdh/ztag113/8735774
• American Heart Association: https://www.heart.org/en/health-topics/heart-attack
• Mayo Clinic: https://www.mayoclinic.org/diseases-conditions/heart-disease/symptoms-causes/syc-20353118
• Cleveland Clinic: https://my.clevelandclinic.org/health/diseases/24129-heart-disease

You Might Like More Articles:

• AI-Powered ECG: How It Works, Benefits & Future of Heart Diagnosis
• Australian Man Lives 105 Days With a BiVACOR Titanium Artificial Heart Before Transplant
• AI Detects Heart Disease Before Symptoms: How AI Is Transforming Early Diagnosis