Impact of voxel size on CT radiomics features: A phantom study / 中国医学影像技术

ABSTRACT

Objective: To investigate the impact of voxel on CT radiomics features. Methods: Catphan 700 phantom was used to perform CT scanning with Siemens definition AS+ 64 row CT scanner on head protocol. The images were reconstructed with different FOV and different thickness, so the ranges of voxel size in these two groups were 0.24-2.38 mm3 and 0.72-2.32 mm3, respectively. ROI was manually sketched using 3D Slicer software, and 108 features were extracted from 7 categories, including shape, first order, gray level dependence matrix, gray level run length matrix, gray level co-occurrence matrix, gray level size zone matrix and neighborhood gray-tone difference matrix. The coefficient of variation (CV) was adapted to evaluate the impact of voxel derived from FOV and slice thickness on CT radiomics features. Results: The voxel had less effect on shape features (CV≤10%), while it had a great impact on most features of other six feature groups (CV>20%). Among them, the dependence entropy, short run emphasis, run entropy, idmn, idn, idm, id and zone entropy were stable (all CV≤10%). Conclusion: The voxel size has a great impact on CT radiomics features. Data preprocessing may be a good way to ensure the stability of features, especially for the application and comparison of multi-centers data.