Livetextbook — a clinical reasoning layer for stroke AI

Livetextbook is an active build. This page describes the concept and current direction; scope, naming, and availability may change. It is navigational guidance, not a medical device and not a diagnostic.

The problem

More data has not automatically produced more confident decisions at the bedside.

Imaging AI has made the quantification of acute stroke better than ever — core, penumbra, occlusion site, ASPECTS, collaterals, all in seconds. But a richer numerical picture of a case does not, on its own, tell the treating physician what the evidence says for this patient, at this hour, with these numbers. That last step — from quantification to decision — is still made by a clinician reasoning over a large, scattered, fast-moving literature. Livetextbook is built for that step.

What it is

Livetextbook is an evidence-organised reasoning layer: clinical knowledge broken into discrete, source-attributed units and linked into the rules and criteria a decision actually turns on. Every assertion carries its provenance back to the primary literature, and the boundaries of what the evidence does and does not support are made explicit. It is designed to sit alongside the imaging-AI tools that produce a case's numerical signature — not to replace clinical judgement, but to make the relevant evidence navigable at the moment it's needed.

A five-stage living evidence cycle surrounding the point of clinical decision — A living corpus: find, verify, structure, apply and re-scan as evidence changes.

Find
Verify
Structure
Apply
Update

From a quantified scan, through the Livetextbook reasoning layer, to a supported clinical decision — Where Livetextbook sits: between the quantified scan and the decision the physician actually makes.

Two tracks, one core

Acute stroke

The flagship track — a high-rigour, fully source-attributed corpus for hyperacute decision-making, designed to complement the imaging-AI tools used in acute stroke workflows.

Outpatient neurology

A breadth-first track covering common outpatient neurology — structured conditions, consult support, and patient-facing material for clinicians and care settings.

Both tracks share one core: a single knowledge schema, one citation discipline, and one authority hierarchy, so quality and provenance are consistent wherever the content is surfaced.

Design principles

Source-attributed by construction. Claims trace back to primary literature; provenance is part of the data, not an afterthought.
Navigational, not prescriptive. It organises evidence for a decision; it does not make the decision or diagnose.
No images, no patient identifiers. Where it interoperates with imaging-AI output, the handoff is numerical only — see the sample below.
Boundaries made explicit. Where evidence is thin or contested, that is stated rather than smoothed over.

Current status

Shared knowledge core and citation discipline defined
Acute-stroke corpus built out to a substantial, structured base
Outpatient-neurology breadth content seeded
Deepening expert-reasoning coverage and evidence-gap analysis
Structured pilots alongside imaging-AI workflows and in clinical settings