Home

Awesome

mnet_deep_cdr

Python version range Code for TMI 2018 "Joint Optic Disc and Cup Segmentation Based on Multi-label Deep Network and Polar Transformation"

Project homepage：http://hzfu.github.io/proj_glaucoma_fundus.html

Install dependencies

pip install -r requirements.txt

Install package

pip install .

OpenCV will need to be installed separately.

  1. The code is based on: TensorFlow 1.14 (with Keras) + Matlab
  2. The deep output is raw segmentation result without ellipse fitting. The Matlab code is the ellipse fitting and CDR calculation (by using PDollar toolbox: https://pdollar.github.io/toolbox/).
  3. You can run the 'Step_3_MNet_test.py' for testing any new image directly.
  4. We also provided the validation and test results on REFUGE dataset in 'REFUGE_result' fold.
  5. Note: Due to the 'scipy.misc.imresize' in SciPy 1.0.0 has been removed in SciPy 1.3.0, the original trained model 'Model_MNet_REFUGE.h5' is not suitable. If you want to segment disc/cup from fundus image, you can consider our newest methods: CE-Net and AG-Net, which obtain the better performances and are also released in:
  6. A pytorch implementation of M-Net could be found in AG-Net: https://github.com/HzFu/AGNet

Main files:

  1. 'Step_1_Disc_Crop.py': The disc detection code for whole funuds image.
  2. 'Step_2_MNet_train.py': The M-Net training code.
  3. 'Step_3_MNet_test.py': The M-Net testing code.
  4. 'Step_4_CDR_output.m': The ellipse fitting for disc and cup, and CDR calculation.

If you use this code, please cite the following papers:

  1. Huazhu Fu, Jun Cheng, Yanwu Xu, Damon Wing Kee Wong, Jiang Liu, and Xiaochun Cao, "Joint Optic Disc and Cup Segmentation Based on Multi-label Deep Network and Polar Transformation", IEEE Transactions on Medical Imaging (TMI), vol. 37, no. 7, pp. 1597–1605, 2018. [PDF]
  2. Huazhu Fu, Jun Cheng, Yanwu Xu, Changqing Zhang, Damon Wing Kee Wong, Jiang Liu, and Xiaochun Cao, "Disc-aware Ensemble Network for Glaucoma Screening from Fundus Image", IEEE Transactions on Medical Imaging (TMI), vol. 37, no. 11, pp. 2493–2501, 2018. [PDF]

There are also some related works for medical image segmentation for your reference:

  1. "Attention Guided Network for Retinal Image Segmentation," in MICCAI, 2019. [PDF] [Github Code]
  2. “CE-Net: Context Encoder Network for 2D Medical Image Segmentation,” IEEE TMI, 2019. [PDF] [Github Code]

Note: for ORIGA and SCES datasets

Unfortunately, the ORIGA and SCES datasets cannot be released due to the clinical policy. But, here is an other glaucoma challenge, Retinal Fundus Glaucoma Challenge (REFUGE), including disc/cup segmentation, glaucoma screening, and localization of Fovea.

License

The code is released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International Public License for NonCommercial use only. Any commercial use should get formal permission first.

Update log: