PedXnet - Official Pytorch Implementation

It is currently incomplete. It will be still updated.

💡 Highlights

Three class-balanced pediatric radiograph datasets (PedXnets) by radiographic views labeling have been constructed.
Robust representation was developed by a radiographic views recognition task on the PedXnets.
Radiographic view representations using Model-PedXNets have been validated by applying them to two pediatric medical tasks, which showed better performance than Model-Baseline, and even equivalent and better performance than Model-ImageNet.

Paper

This repository provides the official implementation code of PedXnet in the following paper:
Supervised representation learning based on various levels of pediatric radiographic views for transfer learning

Authors: Sunggu Kyung, Miso Jang, Seungju Park, Hee Mang Yoon, Gil-Sun Hong, and Namkug Kim

MI2RL LAB

(Under revision...) Nature Scientific Report

Requirements

Linux
CUDA 11.6
Python 3.8.5
Pytorch 1.13.1

📦 PedXnet Framework

1. Clone the repository and install dependencies

$ git clone https://github.com/babbu3682/PedXnet.git
$ cd MTD-GAN/
$ pip install -r requirements.txt

여기부터

2. Preparing data

Download the dataset from GRAZPEDWRI-DX.

The processed dataset directory structure as follows:

data/
    train
        |--  0001_1297860395_01_WRI-L1_M014.png
        |--  0001_1297860435_01_WRI-L2_M014.png
                    .
                    .
                    .
    valid
        |--  0010_0604034401_01_WRI-L1_M014.png
        |--  0010_0604034442_01_WRI-L2_M014.png
                    .
                    .
                    .
    test
        |--  0066_1261433016_01_WRI-L1_M011.png
        |--  0066_1261433167_01_WRI-L2_M011.png
                    .
                    .
                    .

script examples

• train:

CUDA_VISIBLE_DEVICES=4 python -W ignore train.py \
--dataset 'GRAZPEDWRI_DX' \
--train-batch-size 20 \
--valid-batch-size 20 \
--train-num-workers 20 \
--valid-num-workers 20 \
--model 'Downtask_General_Fracture_PedXNet_30Class' \
--loss 'Downtask_GRAZPEDWRI_DX_Loss' \
--optimizer 'adam' \
--scheduler 'poly_lr' \
--epochs 200 \
--warmup-epochs 5 \
--lr 1e-4 \
--min-lr 1e-6 \
--multi-gpu-mode 'Single' \
--device 'cuda' \
--print-freq 10 \
--save-checkpoint-every 1 \
--checkpoint-dir '/workspace/sunggu/3.Child/PedXnet_Code_Factory/checkpoints/Fracture/GRAZPEDWRI_DX_InceptionV3_PedXNet_30Class' \
--save-dir '/workspace/sunggu/3.Child/PedXnet_Code_Factory/predictions/train/Fracture/GRAZPEDWRI_DX_InceptionV3_PedXNet_30Class' \
--memo 'inceptionV3, test, node14'

• test:

CUDA_VISIBLE_DEVICES=2 python -W ignore test.py \
--dataset 'GRAZPEDWRI_DX' \
--test-batch-size 20 \
--test-num-workers 20 \
--model 'Downtask_General_Fracture_PedXNet_30Class' \
--loss 'Downtask_GRAZPEDWRI_DX_Loss' \
--multi-gpu-mode 'Single' \
--device 'cuda' \
--checkpoint-dir '/workspace/sunggu/3.Child/PedXnet_Code_Factory/checkpoints/Fracture/GRAZPEDWRI_DX_InceptionV3_PedXNet_30Class_TEST' \
--save-dir '/workspace/sunggu/3.Child/PedXnet_Code_Factory/predictions/test/Fracture/GRAZPEDWRI_DX_InceptionV3_PedXNet_30Class_TEST' \
--resume "/workspace/sunggu/3.Child/PedXnet_Code_Factory/checkpoints/Fracture/GRAZPEDWRI_DX_InceptionV3_PedXNet_30Class/epoch_?_checkpoint.pth" \
--memo 'inceptionV3, test, node14' \
--epoch ?

🙏 Excuse

We indeed welcome the opportunity to make our dataset and model publicly available, to foster collaboration and further research within the community. However, regarding the dataset, as it comprises medical data, it must receive approval from the Asan Hospital Institutional Review Board (IRB). Therefore, the dataset will be made available on a request-only basis. Please contact us by email by filling out the appropriate form.

📝 Citation

If you use this code for your research, please cite our papers:

⏳ It's scheduled to be uploaded soon.

🤝 Acknowledgement

We acknowledge the open-source libraries, including the Albumentations.

🛡️ License

Project is distributed under MIT License