Under Construction · Planned for Phase II

Phase II — Full WSI Integration

The next evolution of PathoIntern: whole-slide image processing, deep-zoom viewer, risk heatmaps, and UNI foundation model embeddings.

Phase II has not started yet

Phase II development begins after Phase I MVP is completed and validated (estimated Week 7+). The architecture is designed to extend cleanly from Phase I — the same FastAPI backend, PostgreSQL + pgvector database, and Next.js frontend will be extended, not replaced.

What Phase II Will Include

Whole-Slide Image Viewer

OpenSeadragon-powered deep-zoom viewer for interactive exploration of 200,000+ cell slides at full resolution.

Risk Heatmap Overlay

Anomaly score heatmap rendered over the whole slide — pathologist can see exactly which regions the AI flagged.

UNI Foundation Model

4096-dimensional embeddings from the UNI model (Nature Medicine, 2024) trained on 100,000+ pathology slides.

Slide-Level Aggregated Scoring

Patch-level anomaly scores are pooled to a slide-level criticality score with spatial distribution analysis.

Institutional Data Integration

~150 institutional blood smear slides (~200 GB) replace the Kaggle dataset for real-world training and validation.

LIS / EMR Integration

REST API integration with hospital Laboratory Information Systems for automated slide import and report export.

Pathologist Learning Loop

Verdict data from Phase I feeds back into the model fine-tuning pipeline, improving accuracy over time.

Production-Grade Deployment

Cloud-containerized architecture with high-availability database, GPU inference servers, and 99.9% uptime SLA.

Phase I → Phase II: Key Changes

LayerPhase IPhase II
InputPre-extracted patches (PNG/JPG)Whole-slide images (NDPI, SVS, TIFF)
EmbeddingCTransPath · 768-dimUNI · 4096-dim
ViewerStatic image displayOpenSeadragon deep-zoom + heatmap
Scoring scopePer-patch anomaly scoreSlide-level aggregated score
Data sourceKaggle Parasite DatasetInstitutional library (~200 GB)
Model accessPublic weightsHuggingFace access required
InfrastructureDocker Compose (local)Cloud deployment (GPU inference)