README.md

IPN (Image Projection Network): 3D to 2D Image Segmentation in OCTA images

This is an implementation of "Image Projection Network: 3D to 2D Image Segmentation in OCTA Images". IPN is proposed for 3D to 2D segmentation. Our key insight is to build a projection learning module(PLM) which uses a unidirectional pooling layer to conduct effective features selection and dimension reduction concurrently.By combining multiple PLMs, the proposed network can input 3D data and output 2D segmentation results.

Paper

Mingchao Li, Yerui Chen, Zexuan Ji, Keren Xie, Songtao Yuan, Qiang Chen, and Shuo Li. "Image Projection Network: 3D to 2D Image Segmentation in OCTA Images," in IEEE Transactions on Medical Imaging, doi: 10.1109/TMI.2020.2992244.

Requirements

Ubuntu 18.04, Python 3.x, Tensorflow 1.12

Test the code

python test.py

If success, the result will be find in 'logs/test_result/'.

Train for your own data

Data storage structure

Place the data as the following structure.(An example in 'dataset/test').

• dataset
  • train
    • image1(modality 1)
      • name1
        • 1.bmp
        • 2.bmp
        • 3.bmp
        • ...
      • name2
      • ...
    • image2(modality 2)
      • name1
        • 1.bmp
        • 2.bmp
        • 3.bmp
        • ...
      • name2
      • ...
    • ...
    • label
      • name1.bmp
      • name2.bmp
      • ...
  • test(same as train)
  • val(same as train)

Train

python train.py

Change Parameters

Parameters are set in 'options/'

The parameter description is in 'param_help.py'

Model Saver

The best model is saved in 'logs/best_model/'

Other models are saved in 'logs/checkpoints/'

Result

python test.py.

The results will be saved in 'logs/test_result'.

Some examples of our results and corresponding OCTA projection maps are in 'logs/other results/'.