Clinical AI translation

The path from a working research method to something a clinician actually uses — and the stage at which most academic work stops. Two partners in this ecosystem address it from opposite ends: Argus Cognitive as a worked example, and Erik Szabó as the investment-side screen.

“A talented researcher is not an ecosystem.” The bottleneck is translation, not invention.

What the translation stage actually contains

From the Argus case: regulatory documentation, pilot execution, user-interface and workflow refinement, product engineering, and commercial preparation. None of these are research activities, and none happen by themselves.

The five-test screen

Erik Szabó’s project-evaluation framework, applicable directly to speech- and behaviour-based digital biomarkers:

TestThe question it forces
Clinical problemIs there a real, named clinical problem — not a capability looking for a use?
Workflow fitDoes it fit how clinicians actually work, or does it require them to change?
Evidence planIs there a credible, funded plan to generate the evidence?
Regulatory pathIs the MDR route understood, and does the intended use trigger it?
Adoption modelWho pays, and why would they adopt it?

The five dimensions, kept separate

A recurring discipline in this wiki, and the main defence against over-claiming:

  1. Clinical validity — does it measure what it claims?
  2. Clinical utility — does using it change decisions or outcomes?
  3. Intellectual property — who owns what, on what terms?
  4. Regulatory status — what may it be used for, under which framework?
  5. Commercial readiness — could it be sold, supported, adopted?

A claim about any one of these is never a claim about the others. FDA clearance for ReVISION says nothing about Neurora’s regulatory status; a granted patent on Speech Gap says nothing about its clinical utility in therapy monitoring. → Capability readiness

Where this ecosystem currently sits

AssetPosition on the path
Speech GapPatented, sensitivity demonstrated in two languages; prospective validation for the new prognostic use still pending
NIPG capabilitiesMixed — public code through unverified research prototype
Argus / NeuroraProduct stage, an example rather than a component
Multimodal therapy monitoringOpportunity only — no agreed study, no project plan

Erik Szabó · Argus Cognitive · Capability readiness · Precision psychiatry · Funding landscape