Head-and-Neck GTV Segmentation in Multimodal FDG-PET/CT scans

Uses code from this repo for PET-CT data processing, patch based data loading, and patch aggregation

Directory structures for benchmarking experiment

• Basic directory structure for saved_models, model_predictions and model_performances:

  root_dir
      |
      |- hecktor-crFHN_rs113
      |   |
      |   |- msam3d_petct
      |
      |- hecktor-crS_rs113
          |
          |- msam3d_petct
          |- unet3d_petct
          |- unet3d_pet
          |- unet3d_ct
  

• Cross-validation sub-directories:

  • For saved_models and model_predictions, each model directory (ex. unet3d_petct) additionally contains 4 sub-dirs for the crossval splits, like so:

    unet3d_petct
        |
        |- crossval-CHGJ
        |- crossval-CHMR
        |- crossval-CHUM
        |- crossval-CHUS
    

  • In case of model_performances, the scorecard is computed for each model across all the crossval splits. Hence, this root directory doesn't have the crossval sub-dirs

• Late fusion labelmaps and scorecard: Additional model directory named unet3d_latefusion within the hecktor-crS_rs113 directory, in case of model_predictions and model_performances