Unpaired Cross-modal Interaction Learning for COVID-19 Segmentation on Limited CT images [paper]

Data preparation

Download COVID-19-20 dataset.

Preprocess the COVID-19-20 dataset according to the nnUNet.

Download ChestXray14 dataset.

Download ChestXR dataset.

Weight initialization：we follow the extension of UniMiSS for weight initialization. (coming soon)

File directory tree:

├─nnUNet
│  ├─nnUNet_preprocessed
│  │  │  Task115_COVIDSegChallenge # COVID-19-20 dataset
├─Chestxray # ChestXray14 dataset
│  ├─image
│  │  │  xxx.png
│  │  │  ...
├─CXR_Covid-19_Challenge # ChestXR dataset
│  ├─train
│  │  ├─covid
│  │  │  │  cov_xxx.jpg
│  │  │  │  ...
│  │  ├─normal
│  │  │  │  normal-xxx.jpg
│  │  │  │  ...
│  │  ├─pneumonia
│  │  │  │  pneumoniaxxx.jpg
│  │  │  │  ...
│  ├─validation
│  │  ├─covid
│  │  │  │  cov_xxx.png
│  │  │  │  ...
│  │  ├─normal
│  │  │  │  normal_xxx.png
│  │  │  │  ...
│  │  ├─pneumonia
│  │  │  │  pneu_xxx.png
│  │  │  │  ...
├─pretrain_ED # Weight initialization
├─UCI
│  │ *.py
│  │ ...

Train with Chestxray14

python train_withChestXray14.py

Train with ChestXR

python train_withChestXR.py

Evaluate

python evaluate.py