RF-Classification

Segmentation of Stroke Lesions using Random Forest Classification.

Table of Contents

1. About
2. Installation
3. Usage
4. Install dependencies

About

Current implementation of this script is based on Halme, H. et al., 2015.

Basic algorithm

• Feature generation
  1. Z-score normalized voxel intensities
  2. Z-score deviation from global average images
  3. Gaussian smoothing
  4. Local asymmetry
• Classifier training using Random forest (tool: ranger).
• Contextual clustering

Installation

Install Obelix

git clone https://github.com/eimimue/IMIS_CodeSample.git
mkdir build && cd build
cmake ../src
make -j4

Dependencies

(for installation routines, see last section)

• ITK 4.9
• Ranger (Implementation of Random Forests)
• Matlab
• Unix based infrastructure

Usage

Open executeLesionPrediction.m and change the following parameters:

• ICNS_RANDOMFORESTGUMP_BINARY: Path to binary

• ORIGINAL_DATA_PATH: Path to image data

• OUTPUT_PATH: Export folder

• RANGER_BINARY: Path to ranger binary

After the parameters have been set, execute executeLesionPrediction.m.

The image data should be sorted in a file structured as seen below.

DATA_PATH
|
|__ Patient_1
|   |__Modality_1
|   |   Image data
|   |__Modality_2
|   |   Image data
|   |__Modality_3
|   |   Image data
|   |Mask
|   |   Image data
|
|__ Patient_2
|   |__Modality_1
|   |   Image data
|   |__Modality_2
|   |   Image data
|   |__Modality_3
|   |   Image data
|   |Mask
|   |   Image data
|
|__ Patient n
|

Dependencies

ITK 4.9

git clone -b v4.9.0  https://itk.org/ITK.git ITK-4.9.0
cd ITK-4.9.0
mkdir bin && cd bin
cmake ..
make -j4 && make install

Ranger

git clone https://github.com/imbs-hl/ranger.git
cd ranger/source
mkdir build && cd build
cmake ..
make -j4

ISLES Evaluation Script

git clone https://github.com/loli/ISLESevaluation
cd ISLESevaluation
mkdir build && cd build
cmake ..
make -j4