Skip to content

Latest commit

 

History

History
142 lines (101 loc) · 5.63 KB

README.md

File metadata and controls

142 lines (101 loc) · 5.63 KB
Raw

Medical ZOO: Manual

Basics

  • All models accept two parameters: a) the input the channels (in_channels), and b) the segmentation classes (classes) and produce un-normalized outputs
  • All losses accept as input the prediction in 5D shape of [batch,classes,dim_1,dim_2,dim_3] and the target in 4D target shape of [batch, dim_1, dim_2, dim_3]. It is converted to one-hot inside the loss function for consistency reasons. Furthermore the normalization of the predictions is handled here. Dice-based losses return the scalar loss for backward(), and the prediction per channels in numpy to track training progress

Arguments and explanation

  • Arguments that you can modify can be found below:
--batchSz, type=int, default=4, help='The batch size for training and validation'
--dataset_name, type=str, default="iseg2017", choices=('iseg2017','brats2018','brats2019','iseg2019','mrbrains4','mrbrains9','miccai2019')

--dim,  default=(64, 64, 64),  help='The sub-image or sub-volume that you want to crop for 2D specify as dim=(64, 64)' 

--nEpochs, type=int, default=250 ,  help='The training epochs'

--inChannels, type=int, choices=(1,2,3) , help='The desired modalities/channels that you want to use'

--inModalities, type=int, choices=(1,2,3), help='The modalities of the dataset'

--samples_train, type=int, default=10

--samples_val, type=int, default=10

'--split', default=0.9, type=float, help='Select percentage of training data(default: 0.9)')

--lr, default=1e-3, type=float, help='learning rate (default: 1e-3)'

--cuda, default=True, help='whether you want to use cuda'

--model, type=str, default='UNET3D', choices=("RESNET3DVAE",'UNET3D',  'DENSENET1', 'DENSENET2', 'DENSENET3', 'HYPERDENSENET', "SKIPDENSENET3D",
                  "DENSEVOXELNET",'VNET','VNET2')
                  
--log_dir', type=str,     default='../runs/'

--opt', type=str, default='sgd', choices=('sgd', 'adam', 'rmsprop')

Models (more to be added...)

Model # Params (M) MACS (G)
Unet3D 17 M 0.9
Vnet 45 M 12
DenseNet3D 3 M 5.1
SkipDenseNet3D 1.5 M 31
DenseVoxelNet 1.8 M 8
HyperDenseNet 10.4 M 5.8

Supported 3D losses

  • BCE Dice Loss
  • Generalized Dice Loss
  • Dice Loss
  • Weighted Smooth L1 Loss
  • Tags Angular Loss
  • Contrastive Loss
  • Weighted Cross Entropy Loss

Medical Image preprocessing utilities

  • rescale_data_volume(img_numpy, out_dim) : Resize the 3d numpy array to the dim size,:param out_dim is the new 3d tuple
  • transform_coordinate_space(modality_1, modality_2) : Accepts nifty objects. Transfers coordinate space from modality_2 to modality_1
  • normalize_intensity(img_tensor, normalization="mean"): Accepts an image tensor and normalizes it.:param normalization: choices = "max", "mean" , type=str
  • resample_to_output(img_nii, voxel_sizes) : re-samples voxel space

Datasets

In the next version we will provide datasets that will be directly downloaded from google drive for the medical decathlon challenge. Stay tuned.

2018 MICCAI Medical Segmentation Decathlon

Recent official results can be found here.

Task Data Info/ Modalities Train/Test Volume size Classes Dataset size (GB)
1. Brats Multi-modal MRI data (FLAIR, T1w, T1gd,T2w) 484 / 266 - - -
2. Heart Mono-modal MRI 20 / 10 - - -
3. Hippocampus head and body Mono-modal MRI 263 / 131 - - -
4. Liver & Tumor Portal venous phase CT 131 / 70 - - -
5. Lung CT 64 / 32 - - -
6. Pancreas & Tumor Portal venous phase CT 282 / 139 - - -
7. Prostate central gland and peripheral Multi-modal MRI (T2, ADC) 32 / 16 - - -
8. Hepatic vessel & Tumor CT 303 / 140 - - -
9. Spleen CT 41 / 20 - - -
10. Colon CT 41 / 20 - - -

Multi modal brain MRI datasets

Task Data Info/ Modalities Train/Test Volume size Classes Dataset size (GB)
Iseg 2017 T1, T2 10 / 10 - 4 -
Iseg 2019 T1, T2 10 / 30 - 4 -
BraTS 2018 FLAIR, T1w, T1gd,T2w 20 / - - 9 or 4 -
BraTS 2019 FLAIR, T1w, T1gd,T2w 20 / - - 9 or 4 -
MrBrains FLAIR, T1w, T1gd,T2w 20 / - - 9 or 4 -
IXI T1,T2 no labels - - -

2D Medical imaging Datasets

Task Data Info/ Modalities Train/Test Volume size Classes Dataset size (GB)
- - - -
- - - - - -
- - - - -

Supported 3D augmentations

  • Random rotate
  • 3D Elastic deformation
  • Random shift/translate
  • Random scaling-zoom in/out
  • Random crop
  • Random axis flip

License and citation

Advice the LICENSE.md file. For usage of third party libraries and repositories please advise the respective distributed terms. If you want, you can also cite this work as:

@MastersThesis{adaloglou2019MRIsegmentation,
author = {Adaloglou Nikolaos},
title={Deep learning in medical image analysis: a comparative analysis of
multi-modal brain-MRI segmentation with 3D deep neural networks},
school = {University of Patras},
note="\url{https://github.com/black0017/MedicalZooPytorch}",
year = {2019},
organization={Nemertes}}

Support

If you really like this repository and find it useful, please consider (★) starring it, so that it can reach a broader audience of like-minded people. It would be highly appreciated :) !