Image-classification-3d

Radiogenomic image classification of MGMT promoter methylation status in glioblastoma patients

Design

This repository contains code to create a baseline model for 3D image classification using MRI scans. The scripts can be used with several command line arguments to perform relevant experiments. Multiple performance metrics, as well as the model's best train weight, will be available in the Results folder.

Dataset

Please download the dataset from the Kaggle challenge website RSNA-MICCAI Brain Tumor Radiogenomic Classification

Usage

Major packages required

Pytorch
scikit-learn
Monai

Directory structure

code directory (example: exp1)
Input/raw/exp1
Input/prep/exp1
Output/exp1/Results/

Prefil config file before running experiments

Use create_folds.py to split the dataset file into 5 folds.
To obtain the baseline, run train.py with command-line arguments.
Execute train_loop.sh to extend the preceding result to all data folds (five in this case).
Obtain radiomic features using pyradiomics and use do_stat-test.py
To perform feature selection use script RFE_fores.py from command line with a config.json as input.
To obtain prediction result for selected feature use compare_ml_models.py with a config.json file.