Segmentation of brain tissues in MRI image has a number of applications in diagnosis, surgical planning, and treatment of brain abnormalities. However, it is a time-consuming task to be performed by medical experts. In addition to that, it is challenging due to intensity overlap between the different tissues caused by the intensity homogeneity and artifacts inherent toMRI. Tominimize this effect, it was proposed to apply histogram based preprocessing. The goal of this project was to develop a robust and automatic segmentation of WhiteMatter (WM), GrayMatter (GM)) and Cerebrospinal Fluid (CSF) of the human brain.
To tackle the problem, we have proposed Convolutional Neural Network (CNN) based approach and probabilistic Atlas. U-net is one of the most commonly used and best-performing architecture in medical image segmentation, and we have used both 2D and 3D versions. The performance was evaluated using Dice Coefficient (DSC), Hausdorff Distance (HD) and Average Volumetric Difference (AVD).
Once the repository has been clone/downloaded, you have to put your dataset in the following way.
.
├── datasets
│ ├── Training_Set
│ |── Validation_Set
| |── Testing_Set
├── 2D
├── 3D
The code has been tested with the following configuration
- h5py == 2.7.0
- keras == 2.0.2
- nibabel == 2.1.0
- nipype == 0.12.1
- python == 2.7.12
- scipy == 0.19.0
- sckit-image == 0.13.0
- sckit-learn == 0.18.1
- tensorflow == 1.0.1
- tensorflow-gpu == 1.0.1
- Run data_handling_2d_patch.py file to create training and validation .npy files (same holds for 3D)
- Run train_main_2d_patch.py to train your CNN. (same holds for 3D)
