/cnn_fcd_detection

Code for paper 'Convolutional neural networks for automatic detection of focal cortical dysplasia'.

Primary LanguagePython

CNNS for FCD detection

This is the source code for paper "Convolutional neural networks for automatic detection of focal cortical dysplasia".

Motivation

We apply deep learning based methods for the task of automatic detection of FCD.

Dependencies

python >= 3.6
pytorch >= 1.3.0
numpy >= 1.14.0
numba >= 0.43.0
nibabel >= 3.0.2
torchio >= 0.17.0

Also, for preprocessing you will need FreeSurfer software.

Training

Put labeled FCD subjects images into folder data/fcd_brains/. Put masks for them into folder data/masks/. Put control subjects MRI images into folder data/control_brains/.

All MRI images should be in .nii.gz format.

And then simply run shell script with recommended parameters:

./train_script.sh

In parameter.py all tunable parameters are listed.

Examples

Green is side patches. Yellow is middle patches. White is FCD ellipse mask.

Aknowledgements

We acknowledge our project medical partner science-practical center n.a. Pirogov for annotated data. We acknowledge Zhores HPC cluster for the computation facilities.