Model Card for CM-UNet

CM-UNet is a UNet-based model designed for coronary artery segmentation in X-Ray angiography.
It leverages self-supervised pretraining on unannotated datasets and transfer learning on limited annotated data, reducing the need for large-scale manual annotations.

---

Model Details

• Developed by: Camille Challier et al.
• Model type: UNet (convolutional encoder-decoder)
• License: Apache-2.0
• Tasks: Coronary artery segmentation in X-Ray angiography images

Model Sources

• Repository: GitHub
• Paper: arXiv:2507.17779

---

Uses

Direct Use

• Intended for research and educational purposes in medical image segmentation.

---

How to Get Started with the Model

import torch
import numpy as np
import matplotlib.pyplot as plt
from PIL import Image
from UNET.model import UNet

# 1. Load model
model = UNet()  
model.load_state_dict(torch.load("unet_weights.pth", map_location="cpu"))
model.eval()

# 2. Load an image (.npy format)
arr = np.load("example.npy")  # replace with your image path
image = Image.fromarray(arr).resize((256, 256), resample=Image.BICUBIC)
x = torch.from_numpy(np.asarray(image)).unsqueeze(0).float()

# 3. Run inference
with torch.no_grad():
    logits = model(x)

# 4. Postprocess → predicted mask
pred_mask = torch.argmax(logits, dim=1).squeeze(0).numpy()

# 5. Plot input and predicted mask
fig, axs = plt.subplots(1, 2, figsize=(8, 4))
axs[0].imshow(arr, cmap="gray")
axs[0].set_title("Input Image")
axs[0].axis("off")
axs[1].imshow(pred_mask, cmap="gray")
axs[1].set_title("Predicted Mask")
axs[1].axis("off")
plt.show()

---

📖 Citation

If you find this work useful, please consider citing it:

@misc{challier2025cmunetselfsupervisedlearningbasedmodel,
      title={CM-UNet: A Self-Supervised Learning-Based Model for Coronary Artery Segmentation in X-Ray Angiography}, 
      author={Camille Challier and Xiaowu Sun and Thabo Mahendiran and Ortal Senouf and Bernard De Bruyne and Denise Auberson and Olivier Müller and Stephane Fournier and Pascal Frossard and Emmanuel Abbé and Dorina Thanou},
      year={2025},
      eprint={2507.17779},
      archivePrefix={arXiv},
      primaryClass={q-bio.QM},
      url={https://arxiv.org/abs/2507.17779}, 
}