AI interpretation

Interpretation is an optional layer after pipeline outputs exist. Every summary cites computed values from completed runs — the LLM does not invent numbers.

Where to find it

Open /areas/pathology/studies/{id}/interpret directly. This page is not currently linked from the study navigation (Overview / Slides / Cohort / Runs), so bookmark or type the URL until it's added to the sidebar.

Prerequisites

One or more completed pathology pipeline runs (tissue detect, segmentation, spatial quantification, multiplex phenotyping, cohort comparison, or spatial alignment)
AI interpretation enabled for your workspace (ask your administrator if generating an interpretation fails)

Launching interpretation

Select the completed runs you want summarized and generate the interpretation. The page detects when a newer analysis run exists for the same slide/study and flags the existing interpretation as stale.

What you get

A summary of tissue composition and spatial findings
Key findings as bullet points
Hypotheses with cited-metric references back to the source run
Caveats — including quality warnings when tissue or segmentation QC was poor
A research-use-only disclaimer

Guardrails

Every hypothesis cites a specific metric from a completed run — fabricated numbers are rejected before storage
No diagnostic claims or clinical recommendations
Quality caveats surface automatically when upstream QC (tissue fraction, focus score, segmentation confidence) is poor

What interpretation does not do