Radiogenomics

Online Supplement for Prediction of Core Signaling Pathway using Diffusion- and Perfusion-based MRI Radiomics and Next Generation Sequencing in IDH wild type Glioblastoma

Feature Extraction (Radiomics-Code-AMC-Anew)

1. File for feature extraction: Run_test.m

2. Data preparation

    1)  All imaging data should be in Nifty type
    2)  Two imaging data are necessary: segmented mask (our example: reg_mni_ROI.nii) and base image (reg_mni_n4_ss_norm_reg_rs_T2.nii)
    3)  Above two imaging data needs to be co-registered before feature extraction
    4)  For CE-T1w/T1w/T2w/FLAIR image, the base image needs to be white-striped before feature extraction
    5)  Data structure: needs to be given like Figure 1					
    6)  In the root path, only patients’ folders are allowed. Any other file or folder will cause an error
   

3. Code preparation

    1)  Put all the codes within the same folder (name: Radiomics-Code-AMC-Anew)
    2)  You need to download matlab tools: imMinkowski
    3)  You need to download matlab tools: Tools for NIfTI and ANALYZE image (version 1.27)
   

4. How to use

    1)  Unzip the zipped folder (Radiomics-Code-AMC-Anew)
    2)  Open the Matlab software (Run_test.m)
    3)  Addpath to the subfolder (Radiomics-Code-AMC-Anew /imMinkowski)
    4)  Addpath to the subfolder (Radiomics-Code-AMC-Anew /NIfTI_20140122)
    5)  Scale the parameters in the code: modifiable number of bins in histogram analysis and number of adjacent voxels in the texture analysis.
    6)  Set the pixel width and slice thickness
    7)  Set the rootpath (above 2-4) and pathname
    8)  Put the ROI mask and base image name
    9)  Run 
    10) When it properly runs, subject 1, subject 2 .. appears in the background
   

Analysis Pipeline (Analysis-Pipeline-Core Signaling Pathway.R)

1. The R codes are embedded

2. The codes have 4 parts

   1)  Line 5-129: Feature selection via Student's t-test with false discovery rate correction
   2)  Line 134-233: Feature selection via LASSO penalization and calculate AUC for each genetic mutation
   3)  Line 234-329: Feature selection via Random Forest and find top 5 important features
   4)  Line 330- : Calculate diagnostic performance
   

3. The data needs to prepared as csv format

4. Feature data: columns – features, rows- patients

5. Reference data: Name of columns: genetic mutation coded as 1 (positive) or 0 (negative) and name of rows: patients (Figure 2)

Figure 1

Figure 2