-
Notifications
You must be signed in to change notification settings - Fork 102
Radiomics
Radiomics involves extracting scalar features from an image within the region of interest (anaotmical structure). This typically involves accumulating contributions from all the voxels within the region of interest; for example the mean of intensities from all the voxels. Radiomic features can be computed from the original or texture (pre-processed) images. Radiomic & texture features are defined in accordance with the IBSI guidelines.
- First order (intensity)
- Peak and valley
- Shape
- Intensity Volume Histogram (IVH)
- Higher order (texture)
The following demonstrates extraction of radiomic features beginning with CERR's planC format. The parameters used for feature calculation are passed via a JSON format settings file.
% Settings file in JSON format
paramFileName = fullfile(getCERRPath,'ModelImplementationLibrary',...
'RadiomicsModels','Settings','paramsForCtRadiomics.json');
% planC object to load from file
planCFile = fullfile(getCERRPath,'..','Unit_Testing','data_for_cerr_tests',...
'CERR_plans','head_neck_ex1_20may03.mat.bz2');
% Structure name to extract radiomics
strName = 'Target 1';
% Path/name of CSV file to write extracted features
csvFile = 'test_radiomics_features.csv';
% Load planC
planC = loadPlanC(planCFile,tempdir);
planC = updatePlanFields(planC);
planC = quality_assure_planC(planCFile, planC);
% Obtain settings
paramS = getRadiomicsParamTemplate(paramFileName);
% Find matching structure index in planC
indexS = planC{end};
strC = {planC{indexS.structures}.structureName};
structNum = getMatchingIndex(strName,strC,'exact');
scanNum = getStructureAssociatedScan(structNum,planC);
% Call feature calculator
featS = calcGlobalRadiomicsFeatures(scanNum, structNum, paramS, planC);
% Write to CSV
writeFeaturesToCSV(featS,csvFile,{'test'})
This example demonstrates the use of batchExtractRadiomics wrapper function to batch extract radiomics for a patient cohort. The anatomical structure for which features are extracted must be consistently named across the entire cohort and defined in the extraction settings file. If structure is not consistent named, structurte index can be specified via strFileMapC as shown below.
% Directory containing CERR files
dirName = fullfile(getCERRPath,'..','Unit_Testing','data_for_cerr_tests',...
'CERR_plans');
% Settings file containing radiomics settings
paramFileName = fullfile(getCERRPath,'ModelImplementationLibrary',...
'RadiomicsModels','Settings','paramsForCtRadiomics.json');
% File path/name to write the extracted features
csvFile = 'test_batch_radiomics_features.csv';
% Specify structure index per planC file if structure name is not consistent across the patient cohort.
strFileMapC(1,1:2) = {'head_neck_ex1_20may03.mat.bz2',11};
strFileMapC(2,1:2) = {'lung_ex1_20may03',3};
strFileMapC(3,1:2) = {'prostex_9may03',4};
% Call batch feature calculator
featureS = batchExtractRadiomics(dirName,paramFileName,strFileMapC);
% Note: `strFileMapC` is not required if the structure is consistently named and passed instead via JSON
% Write batch to CSV
numIds = length(featureS);
idC = cell(numIds,1);
for iFile = 1:numIds
[~,idC{iFile}] = fileparts(featureS(iFile).fileName);
end
writeFeaturesToCSV(featureS,csvFile,idC)
It is often useful to extract image intensities within the ROI to set the optimal number of discretization levels for feature calculation. This example demonstrates obtaining the scan values from each and every voxel within the region of interest for the cohort of patients.
CERR's radiomics implementation is benchmarked with IBSI, ITK and Pyradiomics.
- Apte AP, Iyer A, Crispin-Ortuzar M, Pandya R, V.van Dijk L, Spezi E, Thor M, Um H, Veeraraghavan H, Oh JH, Shukla-Dave A, Deasy JO (2018). Technical Note: Extension of CERR for computational radiomics: a comprehensive MATLAB platform for reproducible radiomics research, Med. Phys., 45 (8), pp. 3713–3720, August 2018. PMID: 29896896
- Iyer, A., E. LoCastro, H. Veeraraghavan, J. Deasy, and A. Apte (2022). IBSI-Compatible Convolutional Image Texture Filters in CERR, Med. Phys. 49 (6), pp. E686.