README.md

---
license: mit
language:
- en
base_model:
- facebook/dinov2-base
pipeline_tag: image-segmentation
tags:
- micronuclei-segmentation
---

# mnDINO: Accurate and robust segmentation of micronuclei with vision transformer networks
This repository provides the pre-trained mnDINO model for our paper: [mnDINO: Accurate and robust segmentation of micronuclei with vision transformer networks](https://www.biorxiv.org/content/10.64898/2026.03.09.710648v2). The official PyTorch source code is publicly available on [GitHub](https://github.com/CaicedoLab/micronuclei-detection), and the annotated micronuclei dataset can be downloaded through the [Bioimage Archive](https://www.ebi.ac.uk/biostudies/bioimages/studies/S-BIAD2809). 

The mnDINO model is specifically designed for highly efficient and accurate micronuclei segmentation in DNA-stained images across diverse experimental conditions. The model outputs both micronuclei and nuclei segmentation masks simultaneously. To accelerate future research in micronucleus (MN) biology. The dataset, code, and pre-trained model are made publicly available to facilitate future research in micronucleus (MN) biology.
# Usage
### Install Package
```bash
pip install mndino
```

### Load the model
```python
import torch
from mndino import mnmodel
from huggingface_hub import hf_hub_download

model_path = hf_hub_download(repo_id="CaicedoLab/mnDINO", filename="mnDINO_v1.pth")
device = "cuda" if torch.cuda.is_available() else "cpu"
model = mnmodel.MicronucleiModel(device=device)
model.load(model_path)
```

### Make predictions
```python
import skimage
import numpy as np

STEP = 32 # recommended value
PREDICTION_BATCH = 4
THRESHOLD = 0.5

im = skimage.io.imread(your_image_path)
im = np.array((im - np.min(im))/(np.max(im) - np.min(im)), dtype="float32") # normalize image
probabilities = model.predict(im, stride=1, step=STEP, batch_size=PREDICTION_BATCH)

mn_predictions = probabilities[0,:,:] > THRESHOLD
nuclei_predictions = probabilities[1,:,:] > THRESHOLD
```

### Evaluation
```python
import skimage
from mndino import evaluation

mn_gt = skimage.io.imread(your_annotated_image_path)
precision, recall = evaluation.segmentation_report(predictions=mn_predictions, gt=mn_gt, intersection_ratio=0.1, wandb_mode=False)
```