Skip to content

Skycocoo/MS-Lesion-Segmentation

Repository files navigation

Notes

Data

https://portal.fli-iam.irisa.fr/msseg-challenge/data

File details

./

  • [useful] MS_segmentation.ipynb: actual training code (for testing purpose)

  • [useful] Post_processing.ipynb: reconstruct patches into original images, and store the results

  • Test_reconstruct.ipynb: previous Reconstruct file (extracted to class)

  • Test_weight_merge.ipynb: test the idea of weighted reconstruct

  • Test_weight_patch.ipynb: test original reconstruct algo

  • Demonstrate_result.ipynb: plot training results vs ground truth on testing data

  • Format_report.ipynb: format the structure of the network for latex report

  • Test_data.ipynb: test the shape of data

  • Training.py: Training with Dice Loss

  • Training-binary.py: Training with Binary Cross Entropy

  • Training-all.py: Training with all data (not selected)

./model/

  • data.py: fetch data into h5 format, preprocess data for training (padding, patch index, etc.), load data from h5 file
  • dice.py: dice score calculation function
  • generator.py: generate patches for training (different for training / testing dataset)
  • model.py: 3D U-Net model structure setup (Keras)
  • recon.py: reconstruct patches to original image shape (each instance: corresponds to one original image shape); 5 K-fold, 15 images, testing images: 3 (or else storing need additional counter)

About

3D Unet for MS Lesion Segmentation

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published