Awesome

MicroSegNet

Official PyTorch implementation of:

MicroSegNet: A Deep Learning Approach for Prostate Segmentation on Micro-Ultrasound Images (CMIG 2024)

This is the first deep learning model for automated prostate segmentation on micro-ultrasound.

The code is only for research purposes. If you have any questions regarding how to use this code, feel free to contact Hongxu Jiang (hongxu.jiang@medicine.ufl.edu).

Requirements

• Python==3.10.6
• torch==2.1.0
• torchvision==0.16.0
• numpy
• opencv-python
• tqdm
• tensorboard
• tensorboardX
• ml-collections
• medpy
• SimpleITK
• scipy
• pip install -r requirements.txt

Dataset

• Micro-Ultrasound Prostate Segmentation Dataset
• Dataset can be accessed here: https://zenodo.org/records/10475293.

Usage

1. Download Google pre-trained ViT models

• Get models in this link. "imagenet21k/R50+ViT-B_16.npz" is used here.
• Rename your model as: R50-ViT-B_16, ViT-B_16, ViT-L_16.....
• Save your model into folder "model/vit_checkpoint/imagenet21k/".
• If you want to use models pretrained on imagenet21k+imagenet2012, please add configs in "TransUNet/networks/vit_seg_configs.py"

2. Prepare data

• Please go to https://zenodo.org/records/10475293 to download our dataset.
• After downloading, extract the file and put it into folder "data/". The directory structure should be as follows:

.
├── data
│   ├── Micro_Ultrasound_Prostate_Segmentation_Dataset
│   │   ├── train
│   │	└── test
│   └── preprocessing.py
│
├── model
│   └── vit_checkpoint
│       └── imagenet21k
│           ├── R50+ViT-B_16.npz
│           └── *.npz
└── TransUNet

• Run the preprocessing script, which would generate training images in folder "train_png/", data list files in folder "lists/" and data.csv for overview.

python preprocessing.py

• Training images are preprocessed to 224*224 to feed into networks.

3. Train/Test

• Please go to the folder "TransUNet/" and it's ready for you to train and test the model.

python train_MicroUS.py
python test_MicroUS.py

The hard region weight here is set to 4 as default, while you can train models with different weight by specifying it in the command line as follows:

python train_MicroUS.py --weight 10
python test_MicroUS.py --weight 10

References

• Some part of the code is adapted from TransUNet , which provides a very good implementation to start with.
• ViT-pytorch
• segmentation_models.pytorch

Citations

If you use our code or dataset, please cite our paper as below:

@article{jiang2024microsegnet,
  title={MicroSegNet: A deep learning approach for prostate segmentation on micro-ultrasound images},
  author={Jiang, Hongxu and Imran, Muhammad and Muralidharan, Preethika and Patel, Anjali and Pensa, Jake and Liang, Muxuan and Benidir, Tarik and Grajo, Joseph R and Joseph, Jason P and Terry, Russell and others},
  journal={Computerized Medical Imaging and Graphics},
  pages={102326},
  year={2024},
  publisher={Elsevier}
}