Apppublic

KhushbakhtXS/medgemma-pathology-api

sourceHugging Faceupdated 7mo agoView on Hugging Face

0likes

App README

MedGemma Pathology API

A FastAPI backend serving a fine-tuned MedGemma model for H&E stained colorectal histopathology tissue classification.

Model

—Base model: google/medgemma-1.5-4b-it
—LoRA adapter: KhushbakhtXS/medgemma-pathology-lora

Tissue Classes

Code	Description
ADI	Adipose tissue
BACK	Background / slide artifact
DEB	Debris / necrosis
LYM	Lymphocytic infiltrate
MUC	Mucinous component
MUS	Muscularis propria / smooth muscle
NORM	Normal colorectal mucosa
STR	Tumor-associated stroma
TUM	Tumor epithelium (urgent)

Endpoints

Method	Path	Description
GET	`/`	Service info
GET	`/health`	Health check
POST	`/classify`	Classify a histopathology image patch

POST `/classify`

Form fields:

—file (required) — image file (JPEG / PNG)
—clinical_notes (optional) — free-text clinical context

Response:

json

{
  "tissue_type": "TUM",
  "confidence": "high",
  "reasoning": "Irregular glandular structures with nuclear pleomorphism.",
  "guidance": "Tumor epithelium identified. Correlate with grade and staging workup.",
  "urgent": true,
  "raw_output": "..."
}

Configuration

Set the HF_TOKEN Space secret to a Hugging Face token with read access to the gated MedGemma model.