Click here to


Are you sure ?

Yes, do it No, cancel

Evaluation of MRI Radiomics Feature Robustness Using a Virtual Radiomics Phantom

C Ma*, X Wang, K Qing, N Yue, K Nie, Rutgers Cancer Institute of New Jersey, New Brunswick, NJ


(Sunday, 7/12/2020)   [Eastern Time (GMT-4)]

Room: AAPM ePoster Library

Purpose: is recognized that challenges to radiomics include standardization of image acquisition and image processing. Given the bloom of radiomics field, it is crucial to identify robust feature sets and incorporate only those into radiomics-guided clinical trials.The purpose of this study is to evaluate the robustness of MRI radiomics features with various pulse sequence designs through a virtual radiomics phantom.

Methods: digitized human brain phantom were created using a numerical MRI simulator by changing different image sequence parameters. Various T1-weighted, T2-weighted and proton density (PD)-weighted spin echo sequences by changing TR, TE, flip angle, bandwidth etc., were simulated. A total of 64 first-order histogram-based and second-order texture-based radiomics features(Gray Level Co-occurrence Matrix as GLCM, and Gray Level Run Length Matrix as GLRLM) were extracted using an open-source CERR radiomics platform. For each of the image sequences, the mean, standard deviation and coefficient of variation (COV) were calculated for small variations of the TE values.

Results: weighted images demonstrated the highest stability to TE changes in the first-order histogram-based radiomics features, with mean COV of 0.08 among 22 first-order features.T1-weighted and T2-weighted images showed higher sensitivity to TE changes, with Mean COV values of 0.14 and 0.44 respectively in the first-order features.T1-weighted images showed the highest stability in the second-order features, with mean COV of 0.07 among 26 GLCM features and mean COV of 0.08 among 16 GLRLM features. The second-order features showed high sensitivity to TE changes for both PD-weighted and T2-weighted images, with Mean COV values of 0.16 and 0.32 for GLCM, and 0.30 and 0.36 for GLRLM features.

Conclusion: virtual MRI radiomics phantom has been developed using a numerical MRI simulator. It can be used to test the robustness of MRI radiomics features and has the potential to guide in designing reliable image sequences for MRI radiomics-based studies.


MRI, Feature Selection, Pulse Sequences


IM- MRI : Radiomics

Contact Email