Most diagnostic-AI coverage exists as scattered individual claims, a company's press release here, a conference presentation there, a single study covered once and rarely revisited as evidence accumulates or regulatory status changes. What this category lacks, and what this platform intends to build, is a maintained reference: one place tracking clinically relevant diagnostic AI tools against a consistent evidence-ladder framework, updated as the evidence and regulatory picture actually changes rather than frozen at whatever stage existed when a single article was published. This piece publishes the index's structure and methodology first, in the same sequence this platform uses for its other original evidence assets, the standard defined before the data is populated against it, so the eventual index can be held to a commitment made in public rather than one shaped retrospectively around whatever the data happens to show.
What each entry will record
For every tool the index tracks: company and exact product name, never company alone, since regulatory status and evidence attach to specific products rather than to companies as a whole. Specialty and modality, the clinical area and type of input the tool works from. Intended use, stated precisely rather than paraphrased from marketing language. Whether the tool is assistive or autonomous, the distinction this cluster treats as fundamental throughout, since the evidence bar for a tool acting independently is properly higher than for one supporting a clinician's own decision. Regulatory status by territory, since a tool cleared in one jurisdiction may hold a different or no status elsewhere, and status changes over time. Evidence maturity, mapped explicitly against this cluster's eight-rung evidence ladder, technical concept through post-market performance, so a tool's position is stated in comparable terms across entries rather than in each company's own preferred language. Largest validation cohort and whether external validation exists, the size-versus-generalisation distinction this cluster applies to every large-dataset claim. Whether prospective and randomised evidence exists, the rungs that separate a promising retrospective signal from a demonstrated real-world effect. Outcome evidence specifically, whether any study has measured that using the tool changed diagnosis, management or patient outcomes, not merely that the tool discriminates well on its own. Known limitations, stated plainly rather than omitted. Last-reviewed date, because every entry in a category this fast-moving needs a visible currency marker. Primary literature, linked directly rather than summarised without a trail back to the source. And relevant NICE or professional guidance, where a tool's use is governed by a specific national recommendation.
The seven separate scores, not one number
The index deliberately avoids the single overall rating that this category's existing coverage too often defaults to, because collapsing genuinely different properties into one number hides exactly the distinctions this cluster exists to preserve. Technical validation: how thoroughly the underlying model has been tested against data resembling its training population. Clinical validation: whether performance has been confirmed prospectively and externally, on patients and settings beyond the original development data. Workflow evidence: whether the tool has been shown to function well when embedded in an actual clinical process, not just in isolated technical testing. Regulatory maturity: mapped to the precise pathway and territory, never compressed into a bare "approved" or "cleared" label. Transparency: how openly the company publishes its methodology, populations, limitations and ongoing performance data, a dimension most existing coverage ignores entirely despite its direct bearing on how much any other score in the index can be trusted. Equity: whether subgroup performance, across sex, ethnicity, age and relevant clinical variation, has been reported and found consistent, or is simply unknown. And post-market evidence: what, if anything, is known about the tool's performance after deployment, as distinct from its pre-deployment validation studies. A tool can score well on technical and clinical validation while scoring poorly on transparency or equity, and the index is built specifically to let readers see that pattern rather than have it averaged away into a single misleading headline number.
Why this is worth building, and what it will not claim
A constantly updated, evidence-ladder-scored reference is more useful to a working clinician than another one-off product review, because the category's defining problem is not a shortage of individual claims but the absence of anywhere that tracks how those claims age, strengthen or weaken as more evidence accumulates. The index will not claim to certify any device, replace its instructions for use, or substitute for a clinician's own reading of the primary literature for any decision that matters; it is a navigation layer over evidence that continues to belong to its original sources, the same honest boundary this platform applies to every synthesis it produces. Every entry will carry a right-of-reply mechanism for companies whose products are scored, and every score will be revisited on a defined schedule rather than left to calcify at its first-published state, because a tool's evidence ladder position is exactly the kind of fact this category most needs tracked over time rather than fixed in a single article's publication date.
Frequently asked questions
How will companies be able to respond to their entry's scoring?
Through a stated right-of-reply process before and after publication, the same commitment this platform has made for its other original benchmarking work, because a scoring exercise that will not hear correction from the parties being scored has a credibility problem the index is specifically designed to avoid.
Will every tool mentioned across this content cluster eventually appear in the index?
That is the intent, building the index out to cover the full range of products this cluster reviews individually, so that a reader can eventually check a tool's current evidence-ladder position in one place rather than needing to locate and re-read the original article.
Does a high score in this index mean a tool is safe to rely on clinically?
No single score substitutes for clinical judgement on a specific patient; the index exists to make a tool's evidence maturity legible and comparable, not to issue a clearance or endorsement of any kind.