Cancer Imaging Phenomics Toolkit (CaPTk)  1.6.1
Brain Cancer: Glioblastoma Survival Prediction Index

This application provides the survival prediction index (SPI) of de novo glioblastoma patients by using baseline pre-operative multi-parametric MRI analysis [1].

A multivariate model trained on data from the Hospital of the University of Pennsylvania (as described in the study of [1]) is provided with the CaPTk. This can be found in [CaPTk_Installation_Folder]/data/survival.

REQUIREMENTS:

  1. Co-registered Multimodal MRI: T1, T1-Gd, T2, T2-FLAIR, DTI-AX, DTI-FA, DTI-RAD, DTI-TR, DSC-PH, DSC-PSR, DSC-rCBV.
  2. Segmentation labels of the tumor sub-regions: Non-enhancing tumor core (Label=1), Enhancing tumor core (Label=4), Edema (Label=2)
  3. Clinical data: A csv file having patient's demographics (Note that the CSV file should be in ASCII format). Should have age (in first column) and survival (in second column) for training a new model, and age only for survival prediction of new patients.
  4. The data for each patient should be organized in the following directory structure.
    • Subject_ID
      1. features.csv file
      2. CONVENTIONAL
        • "my_T1_file.nii.gz"
        • "my_T2_file.nii.gz"
        • "my_T1CE_file.nii.gz"
        • "my_FLAIR_file.nii.gz"
      3. DTI
        • "my_AX_file.nii.gz"
        • "my_FA_file.nii.gz"
        • "my_RAD_file.nii.gz"
        • "my_TR_file.nii.gz"
      4. PERFUSION
        • "my_rCBV_file.nii.gz"
        • "my_PSR_file.nii.gz"
        • "my_PH_file.nii.gz"
      5. SEGMENTATION
        • "original_segmentation_file.nii.gz"
        • "segmentation_file_in_atlas_space.nii.gz"
  5. The data of multiple patients should be organized in the above mentioned structure and reside under the same folder, e.g.:
    • Data_of_multiple_patients
      1. Subject_ID1
      2. Subject_ID2
      3. ...
      4. Subject_IDn

USAGE:

  • Train New Model:
    1. "Select Subjects". Select the input directory (e.g., Data_of_multiple_patients) that follows the folder structure described above.
    2. "Output". Select the folder where the trained model will be saved.
    3. A pop-up window appears displaying the completion of model building (time depends on the number of patients: ~2*patients minutes).
    4. This application is also available as with a stand-alone CLI for data analysts to build pipelines around, using the following example command:
      SurvivalPredictor.exe -t 0 -i C:/SurvivalInput -o C:/SurvivalModel
      
  • Use Existing Model
    1. "Model Directory". Choose the directory of a saved model.
    2. "Test Subjects". Select the input directory (e.g., Data_of_multiple_patients) that follows the folder structure described above.
    3. "Output". Select the output directory where a .csv file with the SPI for all patients will be saved, and click on 'Confirm'.
    4. A pop-up window appears displaying the completion of SPI calculation. The window will also show the SPI index of the first subject in the Data_of_multiple_patients folder (runtime depends on the number of patients: ~2*patients minutes).
    5. This application is also available as with a stand-alone CLI for data analysts to build pipelines around, using the following example command:
      SurvivalPredictor.exe -t 1 -i C:/SurvivalInput -m C:/SurvivalModel -o C:/SurvivalOutput
      



Reference:

  1. L.Macyszyn, H.Akbari, J.M.Pisapia, X.Da, M.Attiah, V.Pigrish, Y.Bi, S.Pal, R.V.Davuluri, L.Roccograndi, N.Dahmane. M.Martinez-Lage, G.Biros, R.L.Wolf, M.Bilello, D.M.O'Rourke, C.Davatzikos. "Imaging patterns predict patient survival and molecular subtype in glioblastoma via machine learning techniques", Neuro Oncol. 18(3):417-25, 2016. DOI:10.1093/neuonc/nov127