by Rian Huang, Zeyan Xu, Yu Xie, Hong Wu, Zixian Li, Yanfen Cui, Yingwen Huo, Chu Han, Xiaotang Yang*, Zaiyi Liu*, Yi Wang*
This repository is for our ESWA 2023 paper 'Joint-phase Attention Network for Breast Cancer Segmentation in DCE-MRI'.
This repository is based on PyTorch 1.7.1.
-
Clone the repository:
git clone https://github.com/ryandok/JPA.git cd JPA
-
Preprocess data including
CropZeroZone, Resample, Normalization
onC1, C0, BreastPred
of the input. Reference: nnUNet -
Internal structure of
data
folders:${data} [RawData] [DCE-C0] - DCE-C0_0000000.nii.gz - DCE-C0_0000001.nii.gz ... [DCE-C1] - DCE-C1_0000000.nii.gz - DCE-C1_0000001.nii.gz ... [TumorMask] - TumorMask_0000000.nii.gz - TumorMask_0000001.nii.gz ... [Processed] [DCE-C0] - DCE-C0_0000000.nii.gz - DCE-C0_0000001.nii.gz ... [DCE-C1] - DCE-C1_0000000.nii.gz - DCE-C1_0000001.nii.gz ... [BreastPred] - BreastPred_0000000.nii.gz - BreastPred_0000001.nii.gz ... [TumorMask] - TumorMask_0000000.nii.gz - TumorMask_0000001.nii.gz ...
-
Train the model:
cd code python train_JPAnet_3channels_C1&C0&Breast.py --gpu 0 --data_root_path YOUR_PROCESSED_DATA --img1_prefix=DCE-C1 --img2_prefix=DCE-C0 --img3_prefix=BreastPred --label_prefix=TumorMask
-
Test the model:
modified 'utils.image_processing_breastSeg.ImageProcessingBreastSeg' and 'utils.image_processing_tumorSeg.ImageProcessingTumorSegOriginRegion' according to your preprocess methods.
cd code python test_TumorSeg_OriginRes_onBreastPred.py --gpu 0 --data_root_path YOUR_RAW_DATA --img1_prefix=DCE-C1 --img2_prefix=DCE-C0 --label_prefix=TumorMask --backbone=JPAnet_C1&C0&Breast --exp_name=YOUR_EXP_NAME
If JPA is useful for your research, please consider citing:
@article{huang2023joint,
title={Joint-phase attention network for breast cancer segmentation in DCE-MRI},
author={Huang, Rian and Xu, Zeyan and Xie, Yu and Wu, Hong and Li, Zixian and Cui, Yanfen and Huo, Yingwen and Han, Chu and Yang, Xiaotang and Liu, Zaiyi and others},
journal={Expert Systems with Applications},
pages={119962},
year={2023},
publisher={Elsevier}}