README.md

keras tensorflow RMDL

Code for paper 'RMDL: Recalibrated Multi-instance Deep Learning for Whole Slide Gastric Image Classification' accepted by MedIA 2019.

Introduction

This implement is based on GPU (with a minimum memory of 4~5 GB).

This is a Keras (2.2.1) implementation of RMDL-inference with backend of Tensorflow (1.10.1). The code was tested with Anaconda and Python (2.7.15).

    conda install pytorch torchvision cudatoolkit=9.0 -c pytorch

Installation

After installing the dependency:

    pip install pyyaml
    pip install pytz
    pip install tensorboardX==1.4 matplotlib pillow 
    pip install tqdm
    conda install scipy==1.1.0
    conda install -c conda-forge opencv

0. Clone the repo:

   git clone https://github.com/EmmaW8/RMDL.git
   cd RMDL

1. Install dependencies:

   Annoconda environment installation and activation:

   conda create -n tf27 pip python=2.7
   source activate tf27

   Tensorflow installation:

   pip install --ignore-installed --upgrade https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-1.10.1-cp27-none-linux_x86_64.whl

   Keras installation:

    pip install keras==2.2.1

   Install dependencies:

   conda install -c conda-forge opencv
   pip install openslide-python
   pip install numpy==1.14.5
   pip install tqdm
   pip install matplotlib
   pip install scikit-image
   pip install git+https://www.github.com/keras-team/keras-contrib.git

2. Configure your dataset path in configure.yaml with parameter 'data_dir_list'. Download the images and network weights from google drive. you can copy your images (end with '.svs') to the data folder. You can also change the gpu number in gpu_list and define a larger or smaller batch size according to your GPU memory size.

3. Run.
   If you want to run for your self using the provided images, remove the folder Outputs first.

   sh run.sh

   The results will be generated in the Outputs folder.

Citation

@article{wang2019rmdl, title={RMDL: Recalibrated Multi-instance Deep Learning for Whole Slide Gastric Image Classification}, author={Wang, Shujun and Zhu, Yaxi and Yu, Lequan and Chen, Hao and Lin, Huangjing and Wan, Xiangbo and Fan, Xinjuan and Heng, Pheng-Ann}, journal={Medical Image Analysis}, pages={101549}, year={2019}, publisher={Elsevier} }