This is a project by team MediMiners. Team Members: Ganesh, Hasibur, Mizanur, and Sazed (Order is insignificant)

Dataset Link

https://www.kaggle.com/c/rsna-miccai-brain-tumor-radiogenomic-classification/

It contains all the images and train_labels.csv

Environment for Densenet and Data Preprocessing

Please install the required packages using pip install -r requirement.txt in the corresponding conda environment.

DICOM to NIFTII

To convert the DICOM format to NIFTII format, please use dicomtoniftii.ipynb under the preprocessing folder.

Densent 169

For central preprocessing and interval preprocessing, please utilize the corresponding codes. Please select the exact MRI type and provide the accurate data path for smooth execution. The results will be saved in a CSV file for further evaluation and ensembling.

Radiomic Part

the conda environment for the radiomic feature code is radiomic.yml

you can download default vit_h from https://github.com/facebookresearch/segment-anything?tab=readme-ov-file#model-checkpoints

For Radiomic Feature Code run these files sequentially

1. 3dMaskGen.py
   
2. radDataCreation.py
   
3. radiomicClassifier.py
   

Note: ensure that all the data paths in those file is accurate. The code is done by using absolute path

Mask Generation using Segment Anything Model (SAM)

The 3D NIFTII mask for the 3D NIFTII image is generated using script 3dMaskGen.py. Ensure the input directory of the 3D NIFTII images, location of the vit_h of the pretrained SAM and output directory for masks are updated as absolute path has been given there.

Radiomic Feature Creation from 3D NIFTII mask

For volume and energy radiomic feature has been extracted from the 3D mask using the script radDataCreation.py. Ensure the input directory of the 3D NIFTII masks, location of the train_labels.csv and output csv bratsRadiomicData_on_585_samples.csv location is updated.

MGMT value prediction

radiomicClassifier.py script is used to predict the MGMT value from the radbratsRadiomicData_on_585_samples.csv. This script also generates the train and test set and there prediction probabilities.

Ensemble and ROC Curve

The ROC curve is generated differently in the dm_auc_generation.ipynb file. The ROC curve generator takes the predicted values for different models or MRI types, along with their maximum, minimum, and average values, which are provided in the corresponding CSV files in the Prediction_CSV folder.