raeinhashemi / ms-lesion-segmentation

Lesion Segmentation

Config file options

network: "2d-unet", "3d-unet", "2d-dense", "3d-dense", "3d-dense-compact"

loss: "focal_loss", "fbeta_loss", "dice_loss", "categorical_crossentropy", "generalized_dice_loss"

learning_rate: ..., "0.01", "0.005", "0.001", ...

decay_rate: "0.95", "0.9", ...
(decay for learning rate after each 500 epochs)

gpu: "0", "1", "2", "3", "0,1", "0,2", ..., "0,1,2", "0,1,3", ..., "0,1,2,3"
(selection of GPUs to be visible to be program)

dimension_size_X: ..., "64", "128", "256", "512", ...
(dimension sizes of the patches)

final_image_shape: (XXX,XXX,XX)
(shape of the disired resampled images - no effect if "resample" is "false")

sequences: "1", ...
(number of image modalities)

classes: "1", ...
(number of prediction classes)

batch_size: "1", ...
(number of batches for training)

raw_format: "nrrd" or "nii"
(original format of images)

resample: "true" or "false"

patch_wise: "true" or "false"

model_name: "XXXXX.hdf5" or "XXXXX.hd5"
(model name located in the model folder)

Tree structure of files and directories

• root
  • data
    • modified_data (Put your data here, only if resample is False and you don't want to resample)
      • XXX_000.nii
      • XXX_001.nii
      • XXX_002.nii
      • ...
    • prediction (Your predictions will be saved here)
      • XXX_prediction.nii
    • raw_data (Put your data here, only if resample is True and you want to resample)
      • XXX_000.nii
      • XXX_001.nii
      • XXX_002.nii
      • ...
  • lib
    • All (.py) library files (You might have to install some other python packages too, like tensorflow, keras, medpy, etc)
  • model
    • trained models with formats of (XXX.hd5 or XXX.hdf5)
  • logs
• config
• data_import
• model_import
• test.py
• train.py