---
title: MoPET Medical Classification
emoji: 🩺
colorFrom: gray
colorTo: pink
sdk: gradio
sdk_version: 6.22.0
app_file: app.py
short_description: MoPET mixture-of-experts medical image classification
python_version: "3.12"
startup_duration_timeout: 30m
---

# MoPET Medical Classification

This Space demos **MoPET: Parameter-Efficient Mixture-of-Experts for Unified
Medical Image Classification** (EMA4MICCAI 2026 Workshop).

MoPET adapts a *frozen* DINOv3 ViT-B/16 backbone with a learned sparse top-k
router over a heterogeneous pool of LoRA + BOFT PEFT experts injected into the
attention `qkv` projections. A single model consolidates four MedMNIST+
classification tasks (Blood, Breast, Derma, Path) behind one shared,
sparsely-routed expert pool.

## Usage

1. Upload a medical image (blood cell microscopy, breast ultrasound, dermoscopy,
   or colon pathology histology).
2. Select the matching task head.
3. Click **Classify** to get per-class probabilities.

> **Disclaimer:** This is a research artifact, not a diagnostic device. Outputs
> must not be used for clinical diagnosis.

## Links

- [Paper (arXiv)](https://arxiv.org/abs/2607.29462)
- [GitHub](https://github.com/sdoerrich97/mopet)
- [Pretrained Weights](https://huggingface.co/sdoerrich97/mopet_dinov3_unified_blood_breast_derma_path)