Closing the AI Innovation Gap: Harrison.ai’s Path Forward on Radiology AI Regulation
Imagine a radiologist in Sydney, London, or Hong Kong having access to AI tools that can detect hundreds of findings across a chest X-ray and other imaging exams in under 90 seconds and their counterpart in Chicago or Houston working without them. Not because the technology doesn’t exist. Because the regulatory pathway hasn’t kept pace with the science, and patients are the ones waiting.
This innovation gap has real patient impact. Especially in US rural areas, high burdens of chronic disease collide with stretched radiology coverage, meaning chest X-rays taken at night or during surges may not be read until much later and chronic conditions like COPD, lung cancer, and emphysema may be overlooked. A comprehensive AI system that evaluates more than 100 findings at once, mirroring how a radiologist actually reviews an image, can help physicians catch related radiologic abnormalities earlier. For the patient, that can mean the difference between a disease caught at a treatable stage and one found too late.
That gap is real, it is growing, and closing it is why Harrison.ai filed a petition with the US Food and Drug Administration (FDA) in October 2025. The FDA has now completed its review. While the specific mechanism we proposed was not adopted, the petition achieved what mattered most: it put the innovation gap on the public record and forced a conversation the industry needed to have.
The United States faces a projected shortage of up to 42,000 radiologists by 2033. At the same time, AI currently assists with only around 1% of diagnostic radiology tasks in the US, a fraction of what is already routine in the EU, UK, and Asia-Pacific. (see https://www.regulations.gov/document/FDA-2025-P-5560-0001)
This isn’t a technology gap. It’s a regulatory one.
Harrison.ai’s petition focused on a specific category of AI tools: computer-aided detection and/or diagnosis (CAD) and computer aided triage (CADt) software used in radiology. Under our proposal, manufacturers with existing 510(k) clearance could comply with additional guardrails, including post-market monitoring, transparency and training requirements, instead of submitting new 510(k)s for expanded capabilities.
The goal was never to reduce safeguards. It was to get more comprehensive diagnostic support to physicians to better care for their patients, streamlining the regulatory pathway while still appropriately addressing patient safety.
Forty-eight companies, individuals, and organizations submitted public comments on the petition. The range of views was wide. But on one point there was broad agreement: the innovation gap between the US and other jurisdictions is real, patients are bearing its cost, and it demands action.
In its April 2026 response, the FDA acknowledged the substance of the conversation Harrison.ai started: “While FDA disagrees with the partial exemption proposed in the petition…we appreciate the thoughtful proposals reflected in the petition and value the perspectives that it elicited in public comments regarding FDA’s regulation of certain AI-enabled devices.”
We appreciate that acknowledgement. The FDA’s Center for Devices and Radiological Health (CDRH) has worked hard to adapt to the challenges of AI medical technology, including through mechanisms like Predetermined Change Control Plans (PCCPs). A PCCP allows a manufacturer to pre-specify future software updates so that planned improvements can be implemented without filing a new clearance submission each time. The FDA suggested in its response that manufacturers could use PCCPs to achieve some of what our petition sought.
We welcome that constructive spirit. But PCCPs, as currently administered, do not address the core problem. The evidentiary bar for bringing comprehensive, multi-finding AI tools to the US market remains structurally mismatched with how the technology works.
One statistic from the FDA’s own letter makes this point starkly: among all 510(k) submissions filed for devices under a particular CAD product code over several years, every single one was placed on hold in most cases for deficiencies related to performance testing, even where the manufacturer already held a prior CAD clearance. That is not a sign of bad actors. It is a sign that the evidentiary expectations themselves merit a closer look. Every month a comprehensive tool is delayed, patients in underserved communities continue without the diagnostic support that is already routine elsewhere.
Every improvement we pursue is ultimately measured by one thing: whether patients receive better, faster, and safer care. With that in mind, here is where we believe the most productive work lies.
The public comments on our petition surfaced several thoughtful alternatives to our proposal. Three stand out that could apply to traditional 510(k)s and also PCCPs:
– “Consider standalone performance studies as the primary evidence of safety and effectiveness in lieu of MRMC studies.” [1]
Multi‑Reader Multi‑Case (MRMC) studies are resource‑intensive, costly, and often not aligned with real‑world clinical practice. Because clinicians routinely verify and contextualize AI recommendations, well‑designed standalone studies can provide robust, clinically relevant validation, and get effective tools to patients faster, without imposing unnecessary MRMC requirements.
– “Expand acceptance of complementary evidence sources, including well-curated outside-US clinical datasets.” [1]
AI tools deployed across millions of patients in the UK, Australia, and Europe generate rich, real-world performance data demonstrating tangible patient benefit. That evidence should count.
– “Modernise risk–benefit assessment to explicitly recognize clinician efficiency claims (e.g., workflow acceleration, throughput gains, and system-level outcomes) as core clinical benefits alongside device accuracy.” [2]
A tool that helps a radiologist process more studies in less time, while maintaining accuracy, is delivering a clinical benefit. The regulatory framework should reflect that.
There is a systemic issue that deserves more attention. The FDA currently tracks when manufacturers withdraw applications or receive negative decisions. What it does not systematically track is when manufacturers shrink their products, removing findings, narrowing indications, or reducing scope, in response to regulatory feedback. That kind of invisible self-censorship may sometimes be appropriate. But it may also be quietly limiting the tools available to American physicians and patients in ways that are hard to see and harder to measure.
We believe the FDA and the broader stakeholder community would benefit from tracking this. Equally, a formal mechanism for manufacturers to raise concerns about study design requirements before committing to expensive studies, a kind of early-stage “Least Burdensome” flag for pre-submission meetings, would help surface these issues earlier, when they are easier to resolve.
Forty-eight stakeholders weighed in. The problem is now on the public record. And the shared recognition that the system must evolve gives us a foundation to build on.
Harrison.ai will continue to engage with the FDA and other stakeholders to develop safe, practical, evidence-based solutions. With 9 FDA clearances covering 13 chest X-ray and CT brain indications and 3 FDA Breakthrough Designations, we have a demonstrated track record of meeting the FDA’s rigorous evidentiary standards, and we bring that same commitment to quality evidence to the policy conversation. We are committed to doing this collaboratively, and to doing it right.
Because this was never about one petition. It’s about unlocking medical capacity and raising the standard of care with AI for patients in the US, just as much as anywhere else in the world.
[1] Public comment: https://www.regulations.gov/comment/FDA-2025-P-5560-0064
[2] Public comment: https://www.regulations.gov/comment/FDA-2025-P-5560-0065
· Harrison.ai’s original petition summary: Harrison.ai Submits FDA Petition to Increase US Access to Innovative Radiology AI While Maintaining Appropriate Safeguards
· Full petition text: https://www.regulations.gov/document/FDA-2025-P-5560-0001
· FDA’s petition denial letter: https://www.regulations.gov/document/FDA-2025-P-5560-0072
· Harrison.ai’s response to the HHS RFI on AI in clinical care (see Q6 response for additional evidence of the US AI innovation gap): https://www.regulations.gov/comment/HHS-ONC-2026-0001-0207
Harrison.ai is a global healthcare technology company enhancing clinician capacity and patient care through AI automation. Clinician-led and patient-first, our suite of comprehensive solutions supports earlier, more accurate diagnoses and seamlessly integrate into clinical workflows. Harrison.ai solutions are available in 40+ countries and to half of all Australian radiologists. They are clinically deployed at 1,000+ customer sites globally, including 40+ NHS Trusts. Harrison.ai has impacted 6.7 million patients’ lives to date. With a growing footprint across the United States, including 9 FDA clearances covering 13 chest x-ray and CT brain indications, 3 FDA Breakthrough Designations, and NTAP reimbursement, Harrison.ai is committed to supporting innovation, quality, and safety for American healthcare.