Extracting radiomics features for a patient cohort - cerr/CERR GitHub Wiki

1. Import DICOM to CERR

2. Ensure uniform structure names across the cohort.

CERR GUI has a tool to rename structures.

• Open the imported patient in CERR Viewer.
• Type "rename" in the command text box as shown.
• This opens up a new window where the existing structures can be renamed.

3. Define extraction settings via JSON.

• Sample JSON files with basic and advanced extraction settings are provided with CERR and can be modified by the user as required.
• A comprehensive list of available settings for each radiomic feature class, along with descriptions, is available here.

4. CERR allows for calculation of radiomics features from derived images. Sample JSON settings for the supported derived image types are provided below:

• Original image

  "imageType": {
              "Original": {}
               }
  

• Haralick cooccurance

  "imageType": {
  	    "HaralickCooccurance": {
  			          "Type": "All",
                                    "Directionality": 2,
  				  "NumLevels": 128,
  				  "PatchSize": [2, 2, 0]
  		                   }
           }
  

• Law's Textures

  "imageType": {
             "LawsConvolution": {
  			       "Direction": "2d",
                                 "Type" : 3
  		          }
               }  
  

• Wavelets (requires Matlab's Wavelet toolbox)

  
  "imageType": {
  	  "Wavelets": {
  			"Wavelets": "Haar",
  			"Index": "",
  			"Direction": "HLH"
  		      }
  
           }
  

• Gabor

  "imageType":  {
  	  "Gabor": {
  			    "Radius": 2,
                              "Sigma" : 3,
                              "AspectRatio": 1,
                              "Orientation": 45,
                              "Wavlength": 1.5
  		  }
             }
  

• Laplacian of Gaussian

  "imageType": {
  	  "LoG": {
  	              "Sigma_mm": 2,
                        "VoxelSize_mm" : [2, 2, 2]
  		  }
            }
  

• CoLlage (not standardized in IBSI)

  "imageType": {
  	  "CoLlage": {
  			"Dimension": "2d",
                          "Dominant_Dir_Radius": [3, 3, 0],
  			"Cooccur_Radius": [3, 3, 0],
                          "Number_Gray_Levels": 64
  		  }
            }
  

5. Calculate radiomics features.

Sample code to batch-calculate global radiomics features for a patient cohort

% Directory containing CERR files
dirName = 'path/to/dir/containing/cerr/files';

% .json file containing radiomics settings 
% including the names of structures to extract radiomics
paramFileName = 'path/to/file/containing/settings/radiomics_settings.json';

% Excel file path to save features(Optional)
outXlsFile = 'path/to/xls/file';

% Call batch feature calculator
featureS = batchExtractRadiomics(dirName,paramFileName,outXlsFile);