src/BrainIAC/IDHprediction/README.md

IDH Mutation Classification

Overview

We present the IDH mutation classification training and inference code for BrainIAC as a downstream task. The pipeline is trained and infered on T1CE and FLAIR scans, with AUC and F1 as evaluation metric.

Data Requirements

• Input: T1CE and FLAIR MR sequences from a single scan
• Format: NIFTI (.nii.gz)
• Preprocessing: Bias field corrected, registered to standard space, skull stripped
• CSV Structure:

  pat_id,scandate,label
  subject001,scan_sequence,1    # 1 for IDH mutant, 0 for wildtype
  

refer to quickstart.ipynb to find how to preprocess data and generate csv file.

Setup

1. Configuration: change the config.yml file accordingly.

   # config.yml
   data:
     train_csv: "path/to/train.csv"
     val_csv: "path/to/val.csv"
     test_csv: "path/to/test.csv"
     root_dir: "../data/sample/processed"
     collate: 2  # two sequence pipeline  
    
   checkpoints: "./checkpoints/idh_model.00"     # for inference/testing 
   
   train:
    finetune: 'yes'      # yes to finetune the entire model 
    freeze: 'no'         # yes to freeze the resnet backbone 
    weights: ./checkpoints/brainiac.ckpt  # path to brainiac weights
   

2. Training:

   python -m IDHprediction.train_idh
   

3. Inference:

   python -m IDHprediction.infer_idh