More than 1,400 AI-enabled medical devices now carry FDA marketing authorization, and radiology accounts for roughly three-quarters of that total. For radiology IT leaders and clinical department heads evaluating FDA-cleared AI radiology solutions, the landscape in 2026 looks very different from even two years ago: there are cleared tools for nearly every major imaging specialty, the regulatory pathways are better understood, and clinical evidence is accumulating unevenly across categories. What the market still lacks is a clear map of where the evidence is solid, where it is thin, and what the regulatory approval actually means in practice.
This post is that map.
How FDA Clearance Works for AI Radiology Tools
Before reviewing the specialty categories, it helps to understand what FDA clearance actually signals and what it does not.
The 510(k) Pathway
Roughly 96 percent of AI-enabled medical devices reach the market through the 510(k) premarket notification process. A manufacturer submits a 510(k) by demonstrating that their device is substantially equivalent to a legally marketed predicate device. The agency does not require the manufacturer to conduct new clinical trials proving superiority or effectiveness. It requires evidence that the new device performs comparably to the predicate.
For radiology AI, this means that a chest triage algorithm can demonstrate substantial equivalence to an earlier FDA-cleared chest algorithm and receive clearance without generating new prospective trial data. Clearance is a regulatory determination, not a clinical endorsement. The FDA’s guidance on AI in software as a medical device outlines lifecycle expectations for these tools once they are on the market, including requirements for monitoring and transparency.
The median time from submission to 510(k) clearance for AI devices in 2025 was approximately 142 days, with a quarter of devices clearing in under 90 days.
The De Novo Pathway
When a manufacturer believes their device is novel enough that no suitable predicate exists, they can request De Novo classification. This pathway accounts for roughly 3 percent of AI-enabled device authorizations, but it carries more regulatory weight: a successful De Novo creates a new device type category, and that new category then becomes the predicate other manufacturers can cite in future 510(k) submissions.
De Novo decisions effectively set the template for an entire subcategory of AI tools. Several early AI radiology clearances used De Novo precisely because there was no predicate yet for a given task, such as automated triage of a specific finding type on CT.
Understanding the difference matters for procurement. A De Novo-cleared tool was reviewed as a genuinely novel product. A 510(k)-cleared tool demonstrated comparability to an existing device. Neither pathway certifies clinical superiority over current radiologist workflows.
The Specialty Landscape in 2026
Chest X-Ray Triage
Chest X-ray triage is among the most mature categories in FDA-cleared AI. Tools in this category flag critical findings, such as pneumothorax, large pleural effusions, and aortic widening, so that urgent studies reach a radiologist’s worklist first. The AI does not replace the read; it reorders the queue.
This is also the category where the evidence base is most robust. Prospective studies have shown measurable reductions in time-to-read for critical findings when AI triage is active, and the clinical argument is intuitive: getting the right study in front of a radiologist faster produces better outcomes in time-sensitive cases.
Pulmonary nodule management is a closely related subcategory. FDA-cleared algorithms can identify and characterize lung nodules on CT, generate likelihood-of-malignancy scores, and flag cases requiring follow-up. These tools have been widely deployed in lung cancer screening programs, where the workflow benefits are well documented. The medical imaging technology trends shaping 2026 and beyond point toward even tighter integration between AI triage and PACS worklist management.
Breast Imaging
Breast imaging AI has received significant attention, with at least eight FDA-cleared products spanning full-field digital mammography, digital breast tomosynthesis, and MRI. Tools are cleared as assistive or concurrent reading aids, not as autonomous readers.
The evidence in breast imaging is encouraging but still developing. Studies have shown AI systems achieving high area-under-the-curve performance for cancer detection, and combined human-plus-AI reading has demonstrated higher sensitivity than radiologists alone in some controlled settings. The gap, however, is the lack of real-world trial data. Most FDA clearances in this category were supported by retrospective study data rather than prospective randomized trials. A large-scale prospective study is underway in the US to generate more rigorous real-world evidence.
An additional concern is dataset composition. Published analyses have found that most breast imaging AI clearance dossiers did not report race and ethnicity breakdowns for training or validation data. For clinical leaders, this is a meaningful consideration when deploying tools across diverse patient populations.
Stroke Detection and Vascular AI
Stroke triage represents one of the strongest clinical evidence bases among all FDA-cleared AI radiology categories. Several vendors have FDA clearance for large vessel occlusion detection on CT angiography, intracranial hemorrhage identification on non-contrast CT, and ASPECTS scoring for ischemic damage quantification.
The evidence here goes beyond sensitivity and specificity metrics. Peer-reviewed studies using real-world data have documented reduced door-to-groin puncture times in mechanical thrombectomy cases when AI-assisted triage was in use. Faster recanalization translates directly into better functional outcomes in LVO stroke, so the workflow benefit is connected to a hard clinical endpoint.
This is the category where AI in medical imaging diagnostics has moved furthest from theoretical capability to demonstrated clinical utility. It also illustrates an important principle: AI evidence is strongest where speed is most consequential and where the outcome measures are clear.
Fracture Detection
FDA-cleared fracture detection AI has expanded considerably, covering wrist, hip, spine, and rib fractures across several cleared products. These tools are primarily intended as a safety net to flag potentially missed fractures in emergency and high-volume settings, not as primary readers.
Evidence in fracture detection is mixed. Retrospective studies show that AI can identify fractures that were initially missed by radiologists or emergency physicians. Prospective evidence on whether routine deployment changes clinical outcomes, for example, by reducing missed fracture rates at the population level, is still limited. This is a category where cleared status has outpaced real-world outcome data.
The benefits of AI-powered PACS integration are particularly relevant here: fracture AI is most useful when it is embedded in the reporting workflow so that flagged findings reach the radiologist without requiring a separate login or review step.
Cardiac Imaging AI
Cardiac imaging is a smaller but growing slice of FDA-cleared radiology AI. Cleared tools cover echocardiography guidance, automated ejection fraction measurement, coronary artery calcium scoring on non-contrast CT, and myocardial perfusion analysis.
This category is notable for one regulatory development: cardiac imaging AI is the only radiology AI area with dedicated CPT Category I reimbursement codes as of 2026. A first code existed previously; a second was added in January 2026. The existence of established CPT codes gives cardiac AI a meaningful operational advantage over other specialty categories, where facilities must either absorb AI tool costs without reimbursement or pursue individual payer negotiations.
Where the Evidence Is Strongest and Where It Is Thin
To summarize the clinical evidence landscape across categories:
Evidence is strongest in stroke and vascular triage, where multiple multicenter studies have linked AI use to measurable workflow improvements and outcome data. Chest X-ray triage for critical findings also has a solid foundation, particularly for time-sensitive presentations.
Evidence is still developing in breast imaging. The tools are cleared, performance metrics are strong in controlled settings, and clinical benefit is plausible, but prospective real-world trials at scale are not yet complete.
Evidence is thinnest in fracture detection and several other emerging categories, where cleared status was based on retrospective performance analysis without prospective validation.
This pattern matters for budgeting and clinical governance. An FDA-cleared tool is not necessarily a proven tool. AI diagnostics reshaping medical imaging is a real and ongoing process, but the pace of clearance has outpaced rigorous real-world evaluation in several categories.
The Reimbursement Gap
One figure captures the gap between regulatory status and operational reality: of 1,400-plus FDA-cleared AI medical devices, only two have dedicated CPT Category I reimbursement codes, and both are in cardiac imaging. For every other specialty category, AI tool costs are either bundled into facility overhead, negotiated directly with payers, or billed under a limited set of Category III tracking codes with no payment guarantee.
This means that a radiology department deploying FDA-cleared AI for chest triage, fracture detection, or breast imaging is likely operating those tools without a dedicated revenue stream to offset costs. Reimbursement advocacy is active through professional societies, but the timeline for additional CPT codes to mature from Category III to Category I is measured in years.
For radiology IT and department leaders, this is not a reason to avoid AI. It is a reason to be deliberate about which tools provide workflow value that justifies the cost without requiring a reimbursement offset, and which tools depend on reimbursement structures that do not yet exist.
OmniPACS works with facilities at exactly this stage of evaluation, helping radiology teams understand how AI tools integrate with their existing PACS infrastructure before they commit to vendor contracts. The operational costs of deploying a cleared AI tool are often hidden in integration, alert fatigue management, and staff training, not just licensing fees.
What Comes Next: Predetermined Change Control and Adaptive AI
The current regulatory model treats AI devices as largely static after clearance. But AI models improve over time, and retraining a cleared model on new data has traditionally required a new submission. The FDA’s Predetermined Change Control Plan guidance, finalized in late 2024, changes that.
Under PCCP, a manufacturer can define, at the time of clearance, a “playbook” of anticipated modifications, such as retraining on expanded datasets or adding new output types. If the change falls within the approved PCCP, the manufacturer can implement it without a new submission. For radiology AI, this means cleared tools can evolve more responsively as clinical data accumulates, which should accelerate the translation from cleared status to clinical maturity.
The practical implication for facilities is that some cleared AI tools will improve meaningfully over their deployment lifecycle. The tools you deploy today may perform differently in two years, which changes how you should think about vendor relationships, update policies, and ongoing performance monitoring.
OmniPACS supports facilities navigating these infrastructure decisions through a flexible, cloud-based PACS architecture designed to integrate with evolving AI layers without requiring major system replacements. If you are building or updating your imaging infrastructure with AI integration in mind, explore OmniPACS solutions to understand how PACS and AI can be configured to grow together.
A Practical Frame for Radiology Leaders
For a department head or radiology IT leader reviewing the FDA-cleared AI landscape in 2026, the state of the field looks like this:
The volume of cleared tools is large enough that almost any specialty imaging need has at least one cleared option. The quality and evidence base of those options varies significantly by category. Reimbursement exists only in cardiac imaging. Integration burden is often underestimated. And the regulatory framework is evolving to allow more dynamic AI tools over time.
The most defensible approach is to prioritize categories where clinical evidence is already strong, stroke AI and chest triage being the clearest examples, while monitoring breast imaging and fracture detection as evidence matures. For any deployment, the critical question is not whether a tool is FDA-cleared but whether it integrates cleanly into your workflow and whether its performance has been validated on populations similar to yours.
Radiology AI is no longer a future capability. It is a present-tense procurement decision. The tools that will produce the most value are the ones selected for a clear clinical rationale, not just regulatory status. If your imaging infrastructure is not ready to absorb and act on AI outputs, OmniPACS delivers scalable monthly plans to modernize your PACS environment before AI integration becomes the primary constraint.
Frequently Asked Questions
How many FDA-cleared AI radiology tools exist in 2026?
As of early 2026, more than 1,400 AI-enabled medical devices hold FDA marketing authorization across all specialties. Radiology accounts for approximately 76 percent of that total, meaning roughly 1,100 cleared tools are specific to imaging applications. The count includes both diagnostic AI and AI tools that support workflow functions such as triage and worklist prioritization.
What is the difference between 510(k) and De Novo clearance for radiology AI?
510(k) clearance, which accounts for about 96 percent of AI device authorizations, requires a manufacturer to show substantial equivalence to a prior cleared device. No new clinical trial is required. De Novo clearance applies to novel devices with no predicate, requires a more detailed review, and creates a new device category that future 510(k) submissions can reference. A De Novo authorization generally indicates more rigorous regulatory scrutiny than a standard 510(k).
Does FDA clearance mean a radiology AI tool has been clinically proven?
Not necessarily. FDA clearance confirms that a device meets the regulatory standard for marketing, which for 510(k) means substantial equivalence to an existing predicate. It does not require the manufacturer to demonstrate superiority over current clinical workflows or prove patient outcome improvements in prospective trials. Clinical evidence beyond what was required for clearance varies considerably across categories and vendors.
Which radiology AI categories have the strongest clinical evidence?
Stroke triage and large vessel occlusion detection have the strongest evidence base, with multicenter studies documenting measurable workflow improvements and outcome data. Chest X-ray triage for critical findings also has solid supporting research. Breast imaging AI has promising performance metrics but limited prospective trial data. Fracture detection AI has a cleared status with thinner real-world outcome evidence.
Why is reimbursement a challenge for most radiology AI tools?
FDA clearance and insurance reimbursement are separate regulatory processes. As of 2026, only two CPT Category I reimbursement codes exist for radiology AI, both specific to cardiac imaging. Other specialties have limited or no dedicated payment codes, meaning facilities absorb AI costs without a direct reimbursement offset. Payer and professional society advocacy is ongoing to expand the code structure, but the timeline is uncertain.