Y-Net: Joint Segmentation and Classification for Diagnosis of Breast Biopsy Images

This repository contains the source code for our paper, YNet, which is accepted for publication at MICCAI'18.

Sample output of Y-Net

Y-Net identified correctly classified tissues that were not important for diagnosis. For example, stroma was identified as an important tissue, but blood was not. Stroma is an important tissue label for diagnosing breast cancer [1] and removing information about stroma decreased the diagnostic classification accuracy by about 4%. See paper for more details.

[1] Beck, Andrew H., et al. "Systematic analysis of breast cancer morphology uncovers stromal features associated with survival." Science translational medicine 3.108 (2011): 108ra113-108ra113.

Some segmentation results (Left: RGB WSI, Middle: Ground truth, Right: Predictions by Y-Net)

Structure of this repository

YNet is trained in two stages:

stage1 This directory contains the source code for training the stage 1 in Y-Net. Stage 1 is nothing but a segmentation brach.
stage2 This directory contains the source code for training the stage 2 in Y-Net. Stage 2 is jointly learning the segmentation and classification.
seg_eval This directory contains the source code for producing the segmentation masks.

Pre-requisite

To run this code, you need to have following libraries:

OpenCV - We tested our code with version 3.3.0. If you are using other versions, please change the source code accordingly.
PyTorch - We tested with v0.2.0_4. If you are using other versions, please change the source code accordingly.
Python - We tested our code with Python 3.6.2 (Anaconda custom 64-bit). If you are using other Python versions, please feel free to make necessary changes to the code.

We recommend to use Anaconda. We have tested our code on Ubuntu 16.04.

Citation

If Y-Net is useful for your research, then please cite our paper.

@inproceedings{mehta2018ynet,
  title={{Y-Net: Joint Segmentation and Classification for Diagnosis of Breast Biopsy Images}},
  author={Sachin Mehta and Ezgi Mercan and Jamen Bartlett and Donald Weaver and Joann  Elmore and Linda Shapiro},
  booktitle={International Conference on Medical image computing and computer-assisted intervention},
  year={2018},
  organization={Springer}
}

@article{mehta2018espnet,
  title={ESPNet: Efficient Spatial Pyramid of Dilated Convolutions for Semantic Segmentation},
  author={Sachin Mehta, Mohammad Rastegari, Anat Caspi, Linda Shapiro, and Hannaneh Hajishirzi},
  journal={European Conference in Computer Vision (ECCV)},
  year={2018}
}

License