xf4j / brats17 Goto Github PK

Patch-based 3D U-Net for brain tumor segmentation

License: MIT License

Python 100.00%

brats17's Introduction

Synopsis

Submission for Multimodal Brain Tumor Segmentation Challenge 2017 (http://braintumorsegmentation.org/). A patch-based 3D U-Net model is used. Instead of predicting the class label of the center pixel, this model predicts the class label for the entire patch. A sliding-window method is used in deployment with overlaps between patches to average the predictions.

Code Example

The workflow includes bias correction, patch extraction, training, post-processing, testing and submission.

After training data is downloaded, run python bias_correction.py input_dir to perform bias field correction based on N4ITK (https://www.ncbi.nlm.nih.gov/pubmed/20378467). The corrected dataset will be saved at the same folder with the raw dataset.

Run python generate_patches.py input_dir output_dir to generate patches for training.

To train the model, run python main.py --train=True --train_data_dir=train_patch_dir. Or you can modify the default parameters in main.py so that you can just run python main.py. Check model.py for more details about the network structure.

To test the model on validation dataset, run python main.py --train=False --deploy_data_dir=deploy_data_dir --deploy_output_dir=deploy_output_dir. The results will be saved at deploy_output_dir. The network structure for survival prediction is not working good as the result is similar as random guessing. So you can ignore that by setting run_survival to False.

To combine the results and generate the final label maps, run python prepare_for_submission.py input_dir output_dir.

Installation

The model is implemented and tested using python 2.7 and Tensorflow 1.1.0, but python 3 and newer versions of Tensorflow should also work. Other required libraries include: numpy, h5py, skimage, transforms3d, nibabel, scipy, nipype. You also need to install ants for bias correction. Read the instructions for Nipype (http://nipy.org/nipype/0.9.2/interfaces/generated/nipype.interfaces.ants.segmentation.html) and Ants (http://stnava.github.io/ANTs/) for more information.

Contributors

Xue Feng, Department of Biomedical Engineering, University of Virginia [email protected]

brats17's People

Contributors

Stargazers

Watchers

brats17's Issues

Hello, I have installed the nipype 1.0.2 version, but when i run bias_correction.py, it comes that:

IOError: No command "N4BiasFieldCorrection" found on host PC. Please check that the corresponding package is installed.

The latest version of nipype(1.0.2) have already integrated the ants package, and in "nipype.interfaces.ants.segmentation", there also exist "N4BiasFieldCorrection" in it.

Why it still suggest me to check that the corresponding package is installed or not.

Waiting for u reply!
Many thanks!

no model

ModuleNotFoundError Traceback (most recent call last)

---> 10 from model import UNet3D, SurvivalVAE
ModuleNotFoundError: No module named 'model'

Regarding Survival Prediction

The brats2017 website has mentioned you as 3rd winner in the category of survival prediction, then how come you have mentioned that survival prediction task is giving results similar to random guessing?

What is "Instead of predicting the class label of the center pixel"

I found u say "Submission for Multimodal Brain Tumor Segmentation Challenge 2017 (http://braintumorsegmentation.org/). A patch-based 3D U-Net model is used. Instead of predicting the class label of the center pixel, this model predicts the class label for the entire patch. A sliding-window method is used in deployment with overlaps between patches to average the predictions."
What is the "Instead of predicting the class label of the center pixel"?
For example,

In training, if I input a 48x48x48 train patch and a 48x48x48 label patch in one "predicting the class label of the center pixel" network, after some conv layers, it may output a 40x40x40 train patch, so I have to crop label patch to 40x40x40 surround center pixel for input?

and in testing, if I also need to add some background pixels surround the test image? Like the training, I need to add 8 pixel surround the the image?
if add 1 pixel, like :
0 0 0 0 0 0
0 2 2 2 2 0
0 2 2 2 2 0
0 0 0 0 0 0

thanks.

can someone explain how to generate patches?

I can't understand how to generate patches in the code. can someone help me?

something about data

It seems that the download of data is closed now, could u share the url?

list index out of range

Traceback (most recent call last):
File "/home/ubuntu/new/brats17-master/generate_patches.py", line 54, in
pool.map(batch_works, range(n_processes))
File "/home/ubuntu/anaconda3/envs/py2/lib/python2.7/multiprocessing/pool.py", line 253, in map
return self.map_async(func, iterable, chunksize).get()
File "/home/ubuntu/anaconda3/envs/py2/lib/python2.7/multiprocessing/pool.py", line 572, in get
raise self._value
IndexError: list index out of range

hello,when I run python bias_correction.py Brats17TrainingData,the error occured,can you tell me how to solve it?Many thanks!

Weights

Hi, could you share a link to the trained model, or model weights
Thanks

xf4j / brats17 Goto Github PK

brats17's Introduction

Synopsis

Code Example

Installation

Contributors

brats17's People

Contributors

Stargazers

Watchers

Forkers

brats17's Issues

Recommend Projects

Recommend Topics

Recommend Org