Skip to content

Latest commit

 

History

History
54 lines (42 loc) · 1.88 KB

README.md

File metadata and controls

54 lines (42 loc) · 1.88 KB
Raw

lacl

This is a PyTorch implementation of the paper "Lesion-Aware Contrastive Representation Learning for Histopathology Whole Slide Images Analysis".

@inproceedings{li2022lesion,
    author    = {Jun Li, Yushan Zheng*, Kun Wu, Jun Shi, Fengying Xie, Zhiguo Jiang},
    title     = {Lesion-Aware Contrastive Representation Learning For Histopathology Whole Slide Images Analysis},
    booktitle = {Medical Image Computing and Computer Assisted Intervention 
                 -- MICCAI 2022},
    year      = {2022}
}

Our code is modified from repository moco.

Data Preparation

This code use "train.txt" to store the path and pseudo-label of images. An example of "train.txt" file is described as follows:

<path>                         <pseudo-label>
[path to slide1]/0000_0000.jpg 0
[path to slide1]/0000_0001.jpg 0
...
[path to slide2]/0000_0000.jpg 1
...

Note: we assign the pseudo-label for the patches from a WSI as the same of the WSI.

Training

Use "default" contrastive table to train the model by following command. This mode will construct the negative sample pair from all other classes of lesion queue.

python main_lacl.py \
  -a resnet50 \
  --moco-k [number of classes] \
  --mlp --aug-plus --cos \
  --dist-url 'tcp://localhost:10002' --multiprocessing-distributed \
  [your train.txt file folders]

Use "custom" contrastive table to train the model by following command. This mode will construct the negative sample pair from custom settings in lacl/utils.py.

python main_lacl.py \
  -a resnet50 \
  --moco-k [number of classes] --contras-mode 'custom' \
  --mlp --aug-plus --cos \
  --dist-url 'tcp://localhost:10002' --multiprocessing-distributed \
  [your train.txt file folders]