Refined-RadImageNet Model Card

Model Description

Refined-RadImageNet is a collection of pre-trained convolutional neural networks (CNNs) specifically designed for medical imaging applications. These models are trained solely on the RadImageNet dataset, which comprises over 1.35 million annotated CT, MRI, and ultrasound images across various pathologies.

Model Details

Developed by: ConvergingMachine
Model type: Convolutional Neural Networks (ResNet variants)
License: Apache-2.0
Finetuned from model: RadImageNet

Performance Comparison of ResNet Models

This table compares the performance of ResNet models pretrained on 2D RadImageNet using regular and Two2Three convolution techniques across various metrics.

Model	Precision (macro)	Recall (macro)	F1 Score (macro)	Balanced Accuracy	Average Accuracy
ResNet10t	0.4720	0.3848	0.3998	0.3848	0.7981
ResNet18	0.5150	0.4383	0.4545	0.4383	0.8177
ResNet50	0.5563	0.4934	0.5097	0.4934	0.8352

Usage

To utilize the Refined-RadImageNet models in your project, follow these steps:

Install Git LFS (Large File Storage):
```
git lfs install
```
Clone the Repository:
```
git clone https://huggingface.co/ogrenenmakine/Refined-RadImagenet source/
```
This command clones the repository into a directory named source.
Load the Model:

The trained models are implemented using the timm library. Here's how to load a model:
```
import timm
model = timm.create_model('resnet10t', num_classes=165)
```
Replace 'resnet10t' with 'resnet18' or 'resnet50' as needed.

Dataset

The RadImageNet dataset includes medical images of 3 modalities, 11 anatomies, and 165 pathologic labels. It is available upon request at www.radimagenet.com.

Citation

If you use this code or the Refined-RadImageNet models in your research, please cite the following paper:

@inproceedings{yavuz2025policy,
  title={Policy Gradient-Driven Noise Mask},
  author={Yavuz, Mehmet Can and Yang, Yang},
  booktitle={International Conference on Pattern Recognition},
  pages={414--431},
  year={2025},
  organization={Springer}
}

License

This project is licensed under the Apache-2.0 License.

Acknowledgements

Special thanks to the contributors of the Refined-RadImageNet GitHub repository and the developers of the RadImageNet dataset.

Datasets:

convergedmachine
/

Refined-RadImagenet