GBCNet

This is the official repository for the paper titled "Surpassing the Human Accuracy: Detecting Gallbladder Cancer from USG Images with Curriculum Learning" (https://arxiv.org/abs/2204.11433). This paper proposed GBCNet, a specialized CNN model, for classifying gallbladder cancer (GBC) from ultrasound images. GBCNet introduces a novel "multi-scale second-order pooling" block for rich feature encoding from ultrasound images. The paper further proposed a novel visual acuity-based curriculum to train GBCNet. The proposed model beats SOTA deep CNN-based classifiers and human radiologists in classifying GBC from ultrasound images.

Installations

We will update the repository soon with the requirements file for the library installations.

Model Weights

Download the pre-trained models:

Dataset

We contributed the first public dataset of 1255 abdominal ultrasound images collected from 218 patients for gallbladder cancer detection. The dataset can be found at: https://gbc-iitd.github.io/data/gbcu

ROI Detection

The FasterRCNN-based ROI detection model code and weight is available in this link. The output of this model is already stored in the roi_pred.json file in the dataset.

Citation

@inproceedings{basu2022surpassing,
  title={Surpassing the Human Accuracy: Detecting Gallbladder Cancer from USG Images with Curriculum Learning},
  author={Basu, Soumen and Gupta, Mayank and Rana, Pratyaksha and Gupta, Pankaj and Arora, Chetan},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  pages={20886--20896},
  year={2022}
}

License

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License for Noncommercial use only. Any commercial use should obtain formal permission.

Acknowledgement

This code base is built upon Res2Net, MPN-COV, and GSoP. Thanks to the authors of these papers for making their code available for public usage.

sbasu276 / GBCNet