rad_feature_extractor

Feature extraction pipeline using pyradiomics (v 2.1.2).

feature_extractor.py extracts radiomic features from patient scans under PATIENTS_FOLDER and produces a RESULTS_FOLDER with the same folder structure as PATIENTS_FOLDER. Instead of having the dicom files, each day folder in RESULTS_FOLDER will have the combined image (.nrrd file) and a 'Masks' folder containing the generated masks (.mha files). Each patient folder in RESULTS_FOLDER will have the extracted data in csv form with the features as columns and day numbers as rows.

Expected folder structure:

• PATIENTS_FOLDER:

  - Patient1
  
        - Day1
        
        		- RT-struct.dcm
        		- slice1.dcm
        		- slice2.dcm
        		- ...
    	
        - Day2
        
        		- RT-struct.dcm
        		- slice1.dcm
        		- slice2.dcm
        		- ...
    	
        - Day3
        
        		- RT-struct.dcm
        		- slice1.dcm
        		- slice2.dcm
        		- ...
    	
        - ...
        
  - Patient2
  
        - ...
        
  - Patient3
  
        - ...
        
  - ...
  

To modify in code:

1. PATIENTS_FOLDER: folder where each patient folder is located
2. RESULTS_FOLDER: folder where results for each patient will be saved
3. organ: organ of interest ('prostate', 'rectum', 'bladder', etc.) (must be available from the RT-struct)
   • note that the string value for this variable must correspond to an available mask of the same name
   • also, the extraction pipeline might use a defined dictionary in the code that translates the english name of the organ into portuguese.

Default settings of extractor:

The RadiomicsFeaturesExtractor instance uses a resampledPixelSpacing of 0.5cm x 0.5cm x 0.5cm, and normalizes the pixel values of the images. These settings can be modified in the instantiate_extractor() function. For more information, see https://pyradiomics.readthedocs.io/en/latest/radiomics.html. The feature extractor instance also has all available image types enabled.

Additional requirements:

• feature_extractor.py uses plastimatch (http://plastimatch.org/plastimatch.html) to combine the dicom slices and extract the masks from the RT-struct. plastimatch must be installed and added to $PATH.
• the code was written in python 3.6.8

To run:

• python path_to_/feature_extractor.py
• python3 path_to_/feature_extractor.py