Description
Making Pathology AI Inspectable
A prediction is only useful when reviewers can see where it comes from. Smart Pathology Assistant transforms AI-classified tissue patches into an interactive review workflow, mapping predictions directly onto gigapixel prostate biopsy slides so that every result can be examined in its original spatial context.
Key Features
- Deep-zoom slide exploration: Navigate high-resolution whole-slide images from an overview down to individual tissue regions.
- Spatial AI overlays: Display UNI-powered predictions for benign tissue, stroma and Gleason patterns 3–5 as colour-coded regions.
- Region-level evidence: Select any predicted region to inspect its classification, confidence and exact slide location.
- Human-in-the-loop review: Filter tissue classes, control overlay visibility and opacity, and record case review notes.
- Persistent workspaces: Authenticated users can return to their saved slide and overlay settings across sessions.
Behind the viewer is a containerised, multi-service architecture connecting the interactive frontend, application backend, persistent user data and an independent machine-learning inference pipeline. Smart Pathology Assistant turns model output into a transparent decision-support experience designed to assist human review—not replace professional judgement.
To make the experience publicly accessible, we provide a self-contained static edition of the slide viewer using pre-generated predictions. It showcases the same deep-zoom and overlay workflow without requiring deployment of the authenticated backend or computationally intensive UNI inference service; both are part of the complete project.