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.
We will update the repository soon with the requirements file for the library installations.
Download the pre-trained models:
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
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.
@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}
}
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.
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.