Room: AAPM ePoster Library
Purpose: Radiomic features extracted from medical images has a potential to improve diagnostic and prognostic accuracy. Although there are few studies that evaluated the stability of radiomic features, it is still unknow that there is a relationship between radiomic feature stability and each imaging parameter. The purpose of this study was to evaluate stability of radiomic features due to change in scanning conditions and preprocessing for computed tomography (CT) images.
Methods: Chest Phantom with three simulated nodules (CT values: -800 HU, -630 HU, +100 HU) was scanned by various CT acquisition parameters, and radiomic features were extracted. The tube voltage, tube current, and reconstruction kernel were changed, and images were taken 5 times with each parameter. As preprocessing parameters, isotropic voxel size, quantization levels, and wavelet band pass filtering were changed. Coefficient of variance was used to evaluate stability of radiomic features, and radiomic features with the coefficient of variance within 10% were defined stable radiomic features.
Results: The ratio of stable radiomic features with the change of tube voltage, tube current, and reconstruction kernel was 36.2%, 36.4%, and 25.8%, respectively, for simulated nodules of +100 HU. The ratio of stable radiomic features increased as the isotropic voxel size and quantization levels were decreased.
Conclusion: There is a relationship between radiomic feature stability and each imaging parameter. Standardization of imaging parameters might be needed to use radiomic features in clinical situation.