Deep Dense Multi-Path Neural Network For Prostate Segmentation In Magnetic Resonance Imaging (Paper)

This repository contains the code for prostate segmentation in MRI using deep dense multi-path neural network.

Task lists

• Preprocessing
• Training and validation
• Cascade
• Inference on the test set
• Resample the test set predictions to their original spacing
• Usage guideline

About

The Preprocessing and Augmentation pipeline are designed specifically for PROMISE12 dataset. However, they can be easily extended for using with other datasets.

The CNN Architecture proposed in our paper:

Private Dataset

Our model achieved 95.11 DSC on our private dataset when training on both T2 and ADC.

PROMISE12

The PROMISE12 dataset contains heterogenous data from 4 different hospitals and institutes. All images were resampled to the spacing of (0.27, 0.27, 2.2) [x, y ,z]. The training set were split into the training* and validation sets based on the patient meta-data to ensure that both the training* and validation sets contain data from all 4 hospitals/institues. The model achieved an average ~89 DSC on the validation set without further post-processing.

Below are a few examples of the model predictions on the validation set (yellow: groundtruth, red: prediction):

• Case00
  
• Case15
  
• Case25
  

Data Directory

.
├── inputs
│   ├── raw
│   │   ├── test
│   │   │   ├── Case00.mhd
│   │   │   ├── Case00.raw
│   │   │   ├── Case00_segmentation.mhd
│   │   │   ├── Case00_segmentation.raw
│   │   │   ├── ...
│   │   │   ├── Case29.mhd
│   │   │   ├── Case29.raw
│   │   │   ├── Case29_segmentation.mhd
│   │   │   ├── Case29_segmentation.raw
│   │   └── training
│   │   │   ├── Case00.mhd
│   │   │   ├── Case00.raw
│   │   │   ├── Case00_segmentation.mhd
│   │   │   ├── Case00_segmentation.raw
│   │   │   ├── ...
│   │   │   ├── Case49.mhd
│   │   │   ├── Case49.raw
│   │   │   ├── Case49_segmentation.mhd
│   │   │   ├── Case49_segmentation.raw

Dependencies

gluoncv==0.4.0
imageio==2.8.0
imgaug==0.3.0
matplotlib==3.2.0
mxboard==0.1.0
mxnet-cu102mkl==1.6.0
numpy==1.17.4
opencv-python==4.2.0.32
scikit-image==0.16.2
simpleitk==1.2.4

Authours

• Minh Nguyen Nhat To

Citation

If any part of this code is used, please give appropriate citation to our paper.

BibTex entry:

@article{to2018deep,
  title={Deep dense multi-path neural network for prostate segmentation in magnetic resonance imaging},
  author={To, Minh Nguyen Nhat and Vu, Dang Quoc and Turkbey, Baris and Choyke, Peter L and Kwak, Jin Tae},
  journal={International journal of computer assisted radiology and surgery},
  volume={13},
  number={11},
  pages={1687--1696},
  year={2018},
  publisher={Springer}
}

To Be Updated 👍