3DCNNForCRMs

Code for Paper: Automated segmentation of complicated cystic renal masses using 3D V-Net convolutional neural network on MRI

Installing Dependencies

Dependencies can be installed using:

pip install -r requirements.txt

Data Preprocessing

The input images required for model creation, along with data augmentation operations, are conducted offline.

python [certain modality]_process.py

Training

Using the default values for hyper-parameters, the following command can be used to initiate training using PyTorch:

python train_[certain modality].py
--batchSz=2
--nEpochs=500
--classes=2
--inChannels=1
--dropout_rate=0.5
--lr=1e-3
--opt='adamw'
--model='VNET' or 'UNET3D'
--cuda

Validation

You can use the following command to initiate model inference using PyTorch:

python eval_[certain modality].py
--batchSz=1
--classes=2
--inChannels=1
--dropout_rate=0.5
--model='VNET' or 'UNET3D'
--resume='model_saved_path/model.pth'
--cuda

Citation

If this code or article has inspired you, please cite it:

@article{kang2026automated,
  title={Automated Segmentation of Complicated Cystic Renal Masses Using 3D V-Net Convolutional Neural Network on MRI},
  author={Kang, Huanhuan and Jia, Chuang and Wang, Zhongyi and Huang, Bin and Wang, He and Jiang, Jiahui and Liu, Zhe and Cui, Mengqiu and Zhao, Jian and Bai, Xu and others},
  journal={British Journal of Radiology},
  pages={811-822},
  year={2026},
  url={https://doi.org/10.1093/bjr/tqag027},
  publisher={Oxford University Press}
}