FDA clears AI tool that reads MRI to flag Alzheimer's risk
NeuroXT's NEXT-Cue just became a Class II medical device — the latest sign that imaging-grade AI software for neurodegenerative disease is moving from research pitch to regulated clinical tool.
The FDA cleared NeuroXT's NEXT-Cue on October 20, 2026, as a Class II Software as a Medical Device (SaMD). NEXT-Cue reads brain MRI alongside demographic and cognitive-assessment data to support clinicians evaluating whether a patient with mild cognitive impairment (MCI) is likely to progress to Alzheimer's dementia within five years — supplementary information for an individualized care plan, not a standalone diagnosis.
It's a small device clearance with a bigger signal behind it, as Diagnostic Imaging's coverage notes. A month earlier, on September 17, 2026, the FDA issued a final order clarifying the partial 510(k) premarket notification exemption for certain AI/ML-driven radiology CADe/CADx software — the category NEXT-Cue sits inside. Together, the two moves sketch a regulator getting more comfortable with imaging AI as a defined, auditable device class rather than a one-off research tool.
For clinics and device makers building in this space, the practical takeaway is the same every time a new SaMD clears: the imaging pipeline behind the model has to be built for the regulatory record from day one — traceable preprocessing, versioned model outputs, and an architecture an auditor can actually read. It's the kind of groundwork teams working with a custom medical software development partner like YuSMP Group's custom software development team put in place before a clearance submission, not after.
What NEXT-Cue actually does
NeuroXT's AI models are trained to assess Alzheimer's-related pathology primarily from MRI. NEXT-Cue combines that imaging signal with patient demographics and standard cognitive-assessment scores to produce a prognostic read on progression risk for patients already diagnosed with MCI — useful for clinicians deciding how aggressively to monitor, treat, or refer a given patient.
Why the exemption order matters more than it sounds
510(k) clearance is a known path, but the September exemption order changes who needs a full premarket submission for narrower, well-characterized CADe/CADx radiology tools. Fewer AI vendors forced through the heaviest review track means faster iteration on imaging models — and more pressure on the software layer around those models to prove, by design, that outputs are reproducible and traceable.
Shipping imaging software that needs to survive FDA scrutiny? Talk to Epica →
More from the blog.
Designing HIPAA-aware software from day one
Compliance bolted on at the end is expensive and fragile. Here's how we design privacy and auditability into the foundation.
Why your DICOM viewer feels slow (and how we fix it)
The bottleneck is rarely the network. Here's what actually makes medical image viewers crawl — and the engineering that makes them fly.
Building regulated imaging software?
From model pipeline to clinical UI, we build the software layer that holds up under FDA review.