BRAF Subtype Mutation CLassification

This repository contains the implementation of the scan-to-prediction pipeline for non-invasive molecular subtype classification from paper [link here] .

---

Table of Contents

• Installation
• Usage
• Features

---

Segmentation

For the tumor segmentation component, please refer to the repository pLGG_Segmentation. This repository contains the complete implementation of our expert-level pediatric brain tumor MRI segmentation model.

The segmentation pipeline performs two key preprocessing steps:

1. Registration of the T2W MRI to a standard space and skull stripping
2. Segmentation of the tumor from the preprocessed T2W MRI

After running the segmentation pipeline, you will obtain both the preprocessed T2W MRI and its corresponding tumor segmentation mask, which are required for the subsequent classification steps.

To proceed with the molecular subtype classification, install the required environment:

conda env create -f 2d_approach.yml

Usage

The pipeline processes a single T2W nifti image and its corresponding segmentation mask at once. Follow these steps to run the classification:

1. Activate the classification environment:

conda activate 2d_approach

3. Extract tumor slice indices from the segmentation:

## pass the path of the preprocessed T2W MRI and its corresponding segmentation mask
python get_min_max.py --image t2_tr_0000.nii.gz --mask t2_tr_label.nii.gz

4. Prepare the sliced data for classification:

python get_BRAF_data_v2.py

5. Run the classification models:

# Wildtype classifier
python main2.py --saved_model tumor__wildtype_radimagenet_fusion_crosstrain_fullimage_internaltestasvalidationResNet50_imagenet_23_0.73.h5 --subtype wildtype

# Fusion classifier
python main2.py --saved_model tumor_fusion_radimagenet_fullimage_internaltestasvalidationResNet50_imagenet_21_0.75.h5 --subtype fusion

# V600E classifier
python main2.py --saved_model tumor_v600e_radimagenet_wildtypecrosstrain_filteredv600e_fullimage_internaltestasvalidationResNet50__35_0.73.h5 --subtype v600e

6. Generate consensus decision:

python consensus.py

7. Get final classification result:

python decision.py

The final classification output will be printed on the terminal screen along with the running logs.