Program Information
Updates to CERR's Radiomics Toolbox for Speed, Quality Assurance and Clinical Use
A Apte*, A Iyer , J Deasy , Memorial Sloan-Kettering Cancer Center, New York, NY
Presentations
TU-C2-GePD-J(A)-1 (Tuesday, August 1, 2017) 10:00 AM - 10:30 AM Room: Joint Imaging-Therapy ePoster Lounge - A
Purpose: To present updates to CERR’s Radiomics toolbox.
Methods: The updates to the Radiomics toolbox are geared towards making it more useful in clinical research. The major updates include (1) Quality Assurance: Due to the lack of open source tools covering all types of radiomics features and the numerous flavors of such features, reproducibility has been hard to achieve. Professionally engineered software like ITK provides good test coverage with their unit tests. However, such testing does not cover against subtle philosophical differences. Hence, writing tests to compare features derived from two or more different software might be the only way to ensure reproducibility of radiomics features. CERR’s Radiomics toolbox provides tests for all flavors of Haralick and the Run Length features by comparing against the ITK implementation. The tests for radiomics features have been deployed in a continuous integration environment using Jenkins. (2) Clinical Use: MIM’s software’s Matlab API was leveraged to pass raw images to the Radiomics toolbox and to write the derived feature images back to MIM, while providing the users with various options for setting the parameters for generating the feature images. (3) Speed: Patch-wise Haralick texture features are computationally intensive since they involve reducing the neighborhood around each voxel in the region of interest. Since Matlab is inefficient with the use of “for-loops”, we formulated a matrix-based approach for patch-wise Haralick texture calculation. As a result of this approach, the repetitive indexing tasks for the voxels in the region of interest get combined into a single operation. Our formulation resulted in a speedup of the order of 20x over C++ based ITK.
Results: The new features and updates are available as open source, GPL copyright software at https://www.github.com/adityaapte/cerr.
Conclusion: Speed, quality assurance and clinical use make CERR’s Radiomics toolbox ideal for Radiomics research.
Contact Email: