Medical Insights

De Novo: Unlocking Cardiac AI’s Billion Dollar Market Potential

Listen to this article · 6 min listen

The FDA’s De Novo Classification process is the regulatory pathway for any truly novel medical device, setting the evidence bar for new categories. The FDA’s Center for Devices and Radiological Health (CDRH) runs these assessments, and the policies established during Bakul Patel’s time as Director for Digital Health really defined the modern regulatory lens for these products. That framework is what sets the precedent for what kind of validation evidence gets a cardiac AI through the door.

De Novo Classification Framework

When a medical device is genuinely new, meaning it’s low-to-moderate-risk but has no existing “predicate” device to compare it to, the De Novo Classification is its path to market. This is the route most cardiac AI tools with novel functions have to take, and it’s a serious undertaking that typically requires 9-12 months, a stark contrast to the 510(k) pathway that hangs on proving you’re just like something that already exists.

Getting a new classification means you’re not just getting a product cleared. You’re setting a new evidence standard for an entire category of devices. That standard has to include specific clinical validation requirements and also plan for potential failures, a big one being algorithmic bias. The known problem of Algorithmic Bias in ECG Interpretation is a recorded failure mode that shows exactly why the data used to train a model needs to cover a representative population. Any honest review process has to dig into that coverage.

Cardiac AI Validation Evidence

The heart of any regulatory assessment is the cardiac AI validation evidence, which is where a company proves its device is safe and effective. The FDA CDRH is the central repository for this, holding the submission materials from vendors like Cleerly, HeartFlow, and Anumana that connect their cardiac imaging tech to reviewable clinical studies.

  • Cleerly’s AI platform for digital care provides a quantitative analysis of coronary plaque from CCTA scans. Its structural plaque analysis, which adds a safety layer beyond just looking at the degree of stenosis, is backed by the multicenter CERTAIN trial.
  • HeartFlow also focuses on AI for cardiac analysis, and its regulatory materials are a case study in connecting imaging to research. They’ve built a significant patent thicket around their CT-FFR technology.
  • Anumana is another key player with materials linking imaging to studies, and they’re notable for being one of the first ECG-AI technologies to secure its own CPT codes.

The submissions from these companies are effectively defining the evidence standards for market entry in cardiac AI. For anyone building a similar tool, their files are required reading. The FDA SaMD Framework guidance offers more context on the agency’s thinking for software as a medical device.

Clinical Reliability Standards

For cardiac AI, clinical reliability standards ensure the tool performs accurately and consistently over time. The FDA’s Breakthrough Device Designation program can speed up the review for devices that target life-threatening conditions, and cardiology has more of these designations than any other specialty, giving those products a priority lane at the FDA.

Key clinical trials are often supplemented with Real-World Evidence (RWE), which is data pulled from EHRs, patient registries, and insurance claims. This kind of evidence doesn’t just strengthen an FDA submission. It also helps build the economic story for payers by demonstrating value in a messy, real-world patient population. On the development side, Good Machine Learning Practice (GMLP) principles are the guideposts for building safe and effective AI/ML devices. Investors are definitely looking for GMLP compliance during due diligence to see if a company is building something sustainable. GMLP principles for medical devices

Long-term reliability is challenged by algorithmic drift. This is the degradation in performance that happens when the data the model sees in the real world starts to look different from the data it was trained on (e.g., a new scanner is introduced at a hospital). You have to monitor for this drift to sustain performance, and companies need to show they have strong monitoring strategies in place. A company’s quality management system (QMS) is also essential, and the FDA’s Quality Management System Regulation (QMSR) now incorporates the ISO 13485 medical device standard by reference, making it the de facto rulebook.

Review Questions for Imaging Leads

When you’re an imaging service lead reviewing a new technology, you have to ask some hard questions. Your questions should center on the evidence standard that was set when the device got its classification. How deep was the clinical validation? You should also be asking to see the population demographics from the training data, which is the only way to get a real handle on potential algorithmic bias. The instructive read here is that a new classification is a framework decision. The actual evidence behind it is what your team can and should be checking.

So what do you actually do? First, look at the regulatory pathway the device took, does a De Novo make sense for its novelty and risk profile, or did they shoehorn it through a 510(k)? Then, assess the quality of their clinical studies (were they prospective, multi-center, blinded?). Examine the data sources they used for validation, especially looking for real-world data beyond the initial trial. Ask them directly about their strategy for monitoring algorithmic drift and what their approach to GMLP looks like, specifically how they document model changes. Running through these steps gives you a complete evaluation of a cardiac AI platform.

Frequently Asked Questions

What is the significance of the FDA De Novo classification for cardiac AI products?

The De Novo classification is the regulatory pathway for novel medical devices without a predicate, often used by cardiac AI products for genuinely new functions. It establishes the specific evidence standard, including clinical validation requirements, needed for these new device types to gain market entry. This process typically takes 9-12 months to complete.

What kind of evidence is required for cardiac AI validation, particularly regarding algorithmic bias?

Cardiac AI validation evidence must demonstrate a device’s safety and effectiveness, forming the core of regulatory assessment. This includes clinical validation requirements and addressing potential failure modes like algorithmic bias. Reviewers must honestly assess population coverage in model development data to mitigate this bias, which has been recorded in ECG interpretation.

How do clinical reliability standards address long-term performance challenges like algorithmic drift?

Clinical reliability standards ensure consistent and accurate performance, which is paramount for cardiac AI. To address algorithmic drift, which occurs when real-world data distributions shift, companies must demonstrate robust monitoring strategies. Good Machine Learning Practice (GMLP) principles also guide the development of safe and effective AI/ML medical devices, and investors consider GMLP compliance during due diligence.

What specific questions should imaging service leads ask when reviewing cardiac AI technology?

Imaging service leads should inquire about the regulatory pathway taken, assessing its alignment with the device’s novelty and risk. They should evaluate the quality of clinical studies, examine data sources used for validation, and ask about ongoing monitoring for algorithmic drift. Understanding the company’s approach to Good Machine Learning Practice (GMLP) is also crucial for a comprehensive evaluation.

Share
Was this article helpful?

Editorial Team

The Editorial Team at Heart AI Safety Research is a group of writers and editors who cover health, workplace wellness, and everyday care. They research each story and publish under a shared desk byline — experts in their fields, working as one team.