Dilated Saliency U-Net

Tensorflow-Keras Dilated Saliency UNet code DSU-Net implemented to segment White Matter Hyperintensities on brain MR images.

Data Preparation

1. You must make a directory for training and test datasets like the 'data/com_test_configs_2fold_adni60' directory.
2. The directory must contain list files of CSF, FLIAR, IAM, ICV, T1w and WMH(label).

How to Train

Run main.py file to train U-Net, Saliency U-Net or Dilated Saliency U-Net with your choice of data.

The example of each model is described in main.py file.

Train options (e.g. epoch, learning rate ...) can be changed in utils.py file (Please, see the set_parser section).

'--gpu_device' must be set with the available GPU number.

Linux command line example

  $ mkdir results
  $ python3 main.py --gpu_device 2 --depth 1 --num_epochs 80 --fold 1 --lr 1e-5 --reduce_lr_factor 0.5 --img_size 64

Publication

This work has been published in Frontiers in aging neuroscience

Jeong, Yunhee, et al. "Dilated saliency u-net for white matter hyperintensities segmentation using irregularity age map." Frontiers in aging neuroscience 11 (2019): 150.