ULTRA-pytorch

Official Pytorch Code for the paper "ULTRA：Uncertainty-aware Label Distribution Learning for Breast Tumor Cellularity Assessment" , presented at MICCAI 2022.

Introduction

Neoadjuvant therapy (NAT) for breast cancer is a common treatment option in clinical practice. Tumor cellularity (TC), which represents the percentage of invasive tumors in the tumor bed, has been widely used to quantify the response of breast cancer to NAT. Therefore, automatic TC estimation is significant in clinical practice. However, existing state-of-the-art methods usually take it as a TC score regression problem, which ignores the ambiguity of TC labels caused by subjective assessment or multiple raters. In this paper, to efficiently leverage the label ambiguities, we proposed an Uncertainty-aware Label disTRibution leArning (ULTRA) framework for automatic TC estimation. The proposed ULTRA first converted the single-value TC labels to discrete label distributions, which effectively models the ambiguity among all possible TC labels. Furthermore, the network learned TC label distributions by minimizing the Kullback-Leibler (KL) divergence between the predicted and ground-truth TC label distributions, which better supervised the model to leverage the ambiguity of TC labels. Moreover, the ULTRA mimicked the multi-rater fusion process in clinical practice with a multi-branch feature fusion module to further explore the uncertainties of TC labels. We evaluated the ULTRA on the public BreastPathQ dataset. The experimental results demonstrate that the ULTRA outperformed the regression-based methods for a large margin and achieved state-of-the-art results.

Prerequisites:

Python 3.6
Pytorch 1.8
Detailed environment Installation: pip install -r requirements.txt

Downloading datasets:

SPIE-AAPM-NCI BreastPathQ Dataset - Challenge Link | Whole Slide Imaging (WSI) Link

How to Us this Code

Clone this repository:

git clone git@github.com:PerceptionComputingLab/ULTRA.git
cd ULTRA/src

Dataset Preparation

The training, validation and testing folders contain multiple images. The training and validation labels are listed in two csv files.

Data Folder-----
      train----
          99788_1.tif
          99788_2.tif
          .......
      validation----
          99854_1.tif
          99854_2.tif
          .......
      test----
          102174_1.tif
          102174_2.tif
          .......
      train_labels.csv
      val_labels.csv

Training :

(1) Train the backbone network to get the pretrained model.

python train_backbone.py --data_path "Directory contains training and validation data "

(2) Retrain the whole network by loading the pretrained model from the first stage.

python train_ldl.py --data_path "Directory contains training and validation data " --pre_train_path 'pretrain model path'

Testing :

python test.py --checkpoint_path " enter the checkpoint path" --test_path "test dataset directory"

Notice: Do not change the location and the name of the checkpoint path since we would automatically parse the name string to get the model configurations!

Acknowledgement:

Our code is inspired from MUSDL, SSL_CR_Histo, SLEXNet

Citation:

@article{li2022ultra,
  title={ULTRA: Uncertainty-aware Label Distribution Learning for Breast Tumor Cellularity Assessment},
  author={Li, Xiangyu and Liang, Xinjie and Luo, Gongning and Wang, Wei and Wang, Kuanquan and Li, Shuo},
  journal={arXiv preprint arXiv:2206.06623},
  year={2022}
}

JohnleeHIT / ULTRA