On June 4, 2026, the FDA logged 510(k) number K261519 as substantially equivalent: Sonio Suspect, from the Paris-based prenatal-imaging company Sonio. The product-code record reads about as clinically loaded as a regulatory entry can. Product code POK maps to 21 CFR 892.2060, and the FDA's own device classification for that code is “Computer-Assisted Diagnostic Software For Lesions Suspicious For Cancer.” The advisory committee is Radiology, the device class is II, and the clearance type is the part that tells you the most about what actually happened: Special.
Start with the regulation, because it constrains everything else. 892.2060 is the classification FDA created for computer-aided detection and diagnosis software — CADe and CADx — that flags or characterizes findings on a radiological image. It is a Class II generic type with special controls, which is the regulatory category that lets a sponsor reach the market through a 510(k) rather than the far heavier premarket approval route reserved for the highest-risk devices. The classification carries an implicit promise about the device's role: this is adjunctive software. It assists a clinician's read; it does not replace the clinician's judgment. That distinction is the load-bearing wall of the entire clearance.
What “Special” actually means here
The single most informative field in the K261519 record is the clearance type, and it is not Traditional. It is Special. Under FDA's three-track 510(k) framework, a Special 510(k) is reserved for a sponsor modifying its own legally marketed device, where the modification does not change the intended use and does not alter the fundamental scientific technology of the device. In plain terms: Sonio already had a cleared Sonio product on the market, it changed something, and it told the FDA the change was an iteration rather than a reinvention. The Special track exists precisely so that a company refining software it already sells does not have to re-litigate the entire safety-and-effectiveness story from scratch — it has to show the delta is well-characterized and design-controlled.
That is a meaningful signal for anyone tracking the device. A Special 510(k) is not a lower bar in the sense of weaker scrutiny; it is a narrower bar. The agency is reviewing what changed, against the backdrop of a predicate the sponsor itself established. For Sonio, whose existing platform applies machine learning to obstetric ultrasound — the company is known for software that checks whether a fetal anatomy survey captured the views a guideline demands — the Suspect clearance reads as an extension of an established product line rather than a debut. The substantial-equivalence comparison almost certainly runs to a prior Sonio clearance, which is the cleanest possible predicate because the sponsor controls both sides of the comparison.
How the substantial-equivalence logic holds together
Substantial equivalence under section 510(k) is not a claim that a device is good; it is a claim that a device is as safe and effective as a legally marketed predicate, with the same intended use and either the same technological characteristics or different characteristics that do not raise new questions of safety and effectiveness. For a piece of diagnostic software, the technological characteristics that matter are the input data type, the algorithmic approach, the output the clinician sees, and how that output integrates into the diagnostic workflow. When the predicate is the sponsor's own earlier device, the input pipeline and the output presentation are usually shared by design, which collapses most of the “different technological characteristics” analysis into a tractable question: does the modification introduce a new failure mode?
The 892.2060 special controls give the FDA the checklist for answering that. Devices in this category are expected to come with standalone algorithm performance data — sensitivity, specificity, and the operating point on the receiver-operating-characteristic curve — plus, where the indication is for assistance to a reader, a reader study showing that clinicians perform at least as well with the software as without it and ideally better. The special controls also demand rigorous labeling about the intended population, the imaging modality, and the explicit limitation that the output is adjunctive. A Special 510(k) does not waive these controls; it lets the sponsor show that its modified device still satisfies them, leaning on the predicate for the parts that did not change.
There is a naming subtlety worth flagging for precision's sake, the kind of thing this desk exists to catch. The marketed name is “Sonio Suspect,” and the FDA generic name attached to the product code is the broad “software for lesions suspicious for cancer.” The generic classification name describes the regulatory bucket, not the literal clinical task of any individual device that lands in it. The bucket is wide enough to hold a range of CADx tools that characterize image findings as suspicious or not. What the clearance record establishes with certainty is the regulatory frame — Class II, 892.2060, POK, adjunctive, reviewed under radiology controls. The precise clinical indication and the populations studied live in the cleared 510(k) summary itself, and a careful reader should anchor any claim about exactly what Suspect flags to that summary rather than to the generic code name.
Why the pathway is the story
For an IP- and regulatory-strategy audience, the interesting move in K261519 is structural, not clinical. Sonio built a predicate, then used the Special 510(k) lane to iterate on it. That is the playbook for a software-medical-device company that intends to ship improvements on a cadence: establish a Traditional clearance once, then layer modifications through Special 510(k)s when the intended use and core technology hold steady. It keeps the regulatory surface area small with each release and concentrates the heavy evidentiary lift in the original clearance. The cost of that strategy is discipline — the moment a modification touches intended use or swaps the fundamental algorithmic approach, the Special lane closes and the sponsor is back to a Traditional or even a De Novo posture.
The clearance also sits inside a broader regulatory experiment that the 892.2060 classification represents: the agency's attempt to fit adjunctive AI diagnostic software into the existing 510(k) machinery rather than inventing a parallel regime. Every CADx clearance in this code is a data point in how durable that fit is. The recurring tension is the predicate chain — substantial equivalence assumes a stable reference device, but machine-learning models drift and retrain in ways that the original predicate framework was not designed to track. A Special 510(k) like this one is the agency and the sponsor jointly testing whether an iterative software change can be cleanly bounded as “not a new question of safety and effectiveness.” On the evidence of the K261519 record, the answer this time was yes.
The substance to watch, then, is the cleared summary's performance section and its claimed population, because that is where the predicate logic either closes tightly or leaves a seam. The regulatory frame is settled; the clinical envelope is what the document defines.