Zhengyushan / dpathnet

Code for the DRA-Net

DRA-Net: Diagnostic Regions Attention Network for Histopathology WSI Recommendation and Retrieval

This is a PyTorch implementation of the paper DRA-Net:

@Article{zheng2020diagnostic,
  author  = {Zheng, Yushan and Jiang, Zhiguo and Shi, Jun and Xie, Fengying and Zhang, Haopeng and 
             Huai, Jianguo and Cao, Ming and Yang, Xiaomiao},
  title   = {Diagnostic Regions Attention Network (DRA-Net) for Histopathology WSI Recommendation and Retrieval},
  journal = {IEEE Transactions on Medical Imaging},
  year    = {2020},
  doi     = {10.1109/TMI.2020.3046636},
}

It also includes the implementation of the paper DPathNet:

@inproceedings{zheng2020tracing,
	author    = {Zheng, Yushan and Jiang, Zhiguo and Zhang, Haopeng and Xie, Fengying and Shi, Jun},
	title     = {Tracing Diagnosis Paths on Histopathology WSIs for Diagnostically Relevant Case Recommendation},
	booktitle = {Medical Image Computing and Computer-Assisted Intervention},
	year      = {2020},
        pages     = {459--469},
        doi       = {10.1007/978-3-030-59722-1_44},
}

Preprocessing

We need to extract the features of image content under the diagnosis path first.

To extract the features using the CNN trained on the ImageNet dataset, please refer to graph_building.sh

To extract the features using the CNN trained by the pathologists' annotations, please refer to graph_building_with_cnn_training.sh:

Training

To train the DRA-Net, run:

DATADIR = [the directory of the data list generated in the preprocessing step.]

for((FOLD=0;FOLD<5;FOLD++)); 
do
    python main.py --dataset-dir $DATADIR/list_fold_$FOLD --prefix-name DRA-Net\
        --hash-bits 32 --rnn-model GRU --num-rnn-layers 1\
        --num-epochs 300 --batch-size 32 --num-workers 8\
        --dist-url 'tcp://localhost:10001' --multiprocessing-distributed --world-size 1 --rank 0
done

To train the DPathNet, run:

DATADIR = [the directory of the data list generated in the preprocessing step.]

for((FOLD=0;FOLD<5;FOLD++)); 
do
    python main.py --dataset-dir $DATADIR/list_fold_$FOLD --prefix-name DPathNet\
        --hash-bits 32 --rnn-model GRU --num-rnn-layers 1\
        --disable-att --num-epochs 300 --batch-size 32 --num-workers 8\
        --dist-url 'tcp://localhost:10001' --multiprocessing-distributed --world-size 1 --rank 0
done

Data description

The structure of the whole slide image dataset to run the code.

./data                                    # The directory of the data.
├─ 0A00DD22-A08E-4B47-A51B-94A8BD039DAA   # The directory for a slide, which is named by GUID in our dataset.
│  ├─ Large                               # The directory of image tiles in Level 0 (40X lens).
│  │  ├─ 0000_0000.jpg                    # The image tile in Row 0 and Column 0.
│  │  ├─ 0000_0001.jpg                    # The image tile in Row 0 and Column 1.
│  │  └─ ...
│  ├─ Medium                              # The directory of image tiles in Level 1 (20X lens).
│  │  ├─ 0000_0000.jpg
│  │  ├─ 0000_0001.jpg
│  │  └─ ...
│  ├─ Small                               # The directory of image tiles in Level 2 (10X lens).
│  │  ├─ 0000_0000.jpg
│  │  ├─ 0000_0001.jpg
│  │  └─ ...
│  ├─ Overview                            # The directory of image tiles in Level 3 (5X lens).
│  │  ├─ 0000_0000.jpg
│  │  ├─ 0000_0001.jpg
│  │  └─ ...
│  ├─ Overview.jpg                        # The thumbnail of the WSI in Level 3.          
│  ├─ AnnotationMask.png                  # The pixel-wise annotation mask of the WSI in Level 3.
│  └─ BrowsingRecord.pkl                  # The file to store the sequence of browing screens by 
│                                           coordinates ((left, right, top, bottom),...) in Level 3.
├─ 0A003711-3BE4-44E2-9280-89D84E5AF59F
└─ ...