RadFormer
This is the official implementation for the paper "RadFormer: Transformers with global–local attention for interpretable and accurate Gallbladder Cancer detection" https://doi.org/10.1016/j.media.2022.102676
Dataset
To get the Gallbladder Cancer Ultrasound Dataset (GBCU) follow the instructions here.
Model Zoo
- Plesse download the zip file containing models from this link.
- Unzip and save the directory
model_weights
Running the Evaluation Code
- Unzip the model weights, and make sure to keep them in the
model_weights
directory. - Run the following commands.
bash run_test.sh
Running the Training Code
- Unzip the model weights, and make sure to keep them in the
model_weights
directory. - Run the following commands.
bash run_train.sh
For running 10-fold cross-validation, use:
bash run_train_cv.sh
Finetuning on Custom Dataset
- Format the dataset as the GBCU.
- Use the
run_train.sh
scipt, and modify the arguments (--img_dir
,--train_list
,--val_list
) according to the dataset.
Citation
@article{basu2023radformer,
title={RadFormer: Transformers with global--local attention for interpretable and accurate Gallbladder Cancer detection},
author={Basu, Soumen and Gupta, Mayank and Rana, Pratyaksha and Gupta, Pankaj and Arora, Chetan},
journal={Medical Image Analysis},
volume={83},
pages={102676},
year={2023},
publisher={Elsevier}
}