De-identify first
Remove names, medical record numbers, dates of birth, addresses, phone numbers and email addresses before entering text or uploading images. Check filenames and any text burned into image labels.
Evidence-grounded pathology workspace
De-identified intake · literature-grounded analysis · structured consultation
Consultation intake
Structured intake with evidence-restricted diagnostic support.
Organ and site, specimen type, age and sex if relevant, clinical history, imaging, prior pathology or treatment.
Architecture, growth pattern, cytology, stroma, mitoses, necrosis, invasion, inflammation, vascular pattern.
Low and high power H&E, IHC, special stains, gross or radiologic images. Images are downscaled and stripped of metadata in your browser before upload. Each image is reduced to 1,568 pixels on the long edge, so include a high power field wherever nuclear or cytologic detail matters. Crop out slide labels.
Record results under each category. Pattern detail materially changes interpretation.
FISH, NGS, RNA fusion, cytogenetics, flow cytometry, microbiology. Include pending results.
The specific uncertainty or distinction you want addressed. Naming the entities in play produces sharper evidence retrieval.
Synthesis of clinical context, morphology, images, immunophenotype and ancillary findings.
Three to four diagnoses with confidence, supporting findings, findings against and the key discriminator.
Roughly three less likely entities with brief rationale.
Smallest high-yield panel with purpose, expected pattern and pitfalls.
Favoured diagnosis where supportable, residual uncertainty and the most efficient next step.
Findings are never invented. Where evidence is insufficient the consultation says so rather than forcing a diagnosis.
Diagnostic evidence is retrieved from indexed peer-reviewed literature and ranked for diagnostic relevance. Authoritative classification frameworks are surfaced as directory references.
Workspace
Drafts and submitted consultations, synced to this workspace.
Case overview
Starting points
Pre-structured intakes you can adapt. Replace every field with your own de-identified findings.
Policy
What the consultation engine may and may not use as diagnostic evidence.
Each case is parsed into anatomic site, immunophenotype, molecular findings, morphologic pattern and named entities. Several complementary queries are then run against indexed peer-reviewed literature — one each for immunophenotype interpretation, morphologic differential, classification and reporting, the consultation question, molecular correlation, and foundational references.
Results are gated on the organ system and on query-specific anchor terms, scored on weighted topical fit, then balanced so that a single query intent or a run of case reports cannot fill the evidence set. Journal standing, recency and citation count act only as tie-breakers between passages that are already on topic.
The consultation model receives these passages with identifiers and may cite them only by identifier. The server resolves identifiers back into references it already holds, so a fabricated citation cannot reach you.
These are licence-restricted and are surfaced as directory references for you to consult directly. They are never reproduced as evidence text.
For tumour classification and diagnostic criteria: current WHO Blue Books and expert diagnostic references, then organ-specific consensus guidelines, then high-quality peer-reviewed literature, then supportive references.
For reporting, staging and biomarkers: CAP, ICCR and AJCC take priority within their scope.
When sources conflict, the disagreement is identified and the newest authoritative classification applicable to the question is favoured, with the uncertainty stated rather than silently resolved.
General medical blogs, patient-facing sites, forums, social media, encyclopaedias, commercial SEO pages, unsourced AI-generated material and non-peer-reviewed diagnostic claims are not used as evidence for the differential or the recommended workup.
Authorised access does not grant redistribution rights. Subscription and copyrighted resources are integrated only in accordance with their licences. The application synthesises and cites evidence rather than reproducing substantial copyrighted text.
Guidance
Four habits that materially improve consultation quality.
Remove names, medical record numbers, dates of birth, addresses, phone numbers and email addresses before entering text or uploading images. Check filenames and any text burned into image labels.
Enter clinical context, morphology, stain results and ancillary findings as observed. A finding you omit is treated as not assessed, not as negative — so state genuine negatives explicitly.
A question that names the diagnoses you are weighing retrieves far sharper evidence than a general request. "Distinguish A from B" outperforms "what is this?".
Every claim that depends on literature is linked to a retrieved passage. Preview the evidence before submitting, and check the citations attached to any conclusion you intend to act on.
Configuration
Appearance, workspace behaviour and backend connections.
Store it in a password manager. If you lose it, the server copies of those cases cannot be recovered.
Before you submit
This interface is built around de-identified case consultation.
Before submitting, verify that case text, image content, filenames and embedded image metadata contain no names, medical record numbers, dates of birth, addresses, contact details or other direct identifiers.
The intake form scans for common identifier patterns — record-number phrasing, date formats, national identifier formats, phone numbers and email addresses — and warns you. This is a convenience check with real limits: it will miss identifiers it does not recognise, and it is not a substitute for your organisation's privacy review.
Images are downscaled and re-encoded in your browser before leaving the device. Re-encoding discards EXIF and other embedded metadata, including acquisition details and any geolocation. It does not remove identifiers that are visually burned into the image itself — check slide labels and annotations yourself.
Server-side case storage uses Netlify Blobs, which is not covered by Netlify's HIPAA-compliant hosting offering. Use this deployment for de-identified records only. Storing protected health information requires a separately validated compliant architecture and appropriate agreements in place.
About
A consultation workflow, not a diagnostic authority.
The workflow organises case context, morphology, images, immunohistochemistry, ancillary testing and the diagnostic question into a consistent intake, retrieves relevant peer-reviewed literature, and returns a structured consultation that cites that evidence, with each citation checked against its claim.
It does not replace pathologist judgement. The connected model is constrained to preserve uncertainty, distinguish observation from inference, cite only supplied evidence and avoid inventing findings.
The mark combines a microscope with connected nodes — pathology expertise joined to computational analysis. It appears in the product header, the loading sequence and the site favicon.
Consultation
Submitted case