Robustness of radiomics features of virtual unenhanced and virtual monoenergetic images in dual-energy CT among different imaging platforms and potential role of CT number variability

Insights into Imaging

Table 2 Inter-platform reproducibility of radiomics among all platforms within the VUE images and within the VMI_70keV

Feature class	CV < 10%		CV mean		QCD < 10%		QCD mean
Feature class	VUE (%)	VMI_70keV (%)	VUE	VMI_70keV	VUE (%)	VMI_70keV (%)	VUE	VMI_70keV
First order (19 features)	8.88	26.64	0.5007	0.7358	44.74	50.99	0.4037	0.2484
Texture (75 features)	19.83	20.58	0.4232	0.4042	41.25	38.67	0.2791	0.2625
GLCM (24 features)	16.93	16.15	0.5460	0.4878	30.21	28.91	0.4073	0.3402
GLDM (14 features)	20.54	20.54	0.3081	0.3151	30.36	29.46	0.1839	0.1923
GLRLM (16 features)	7.03	7.03	0.3716	0.3602	25.39	21.48	0.2156	0.2244
GLSZM (16 features)	5.47	8.20	0.3975	0.3966	14.84	15.23	0.2482	0.2558
NGTDM (5 features)	0.00	0.00	0.4024	0.4176	5.00	12.50	0.2328	0.2299
Overall (94 features)	11.30	15.16	0.4388	0.4712	28.39	28.99	0.3043	0.2597

Percentage indicates the percentage of features met the cutoffs for robustness measures (CV < 10% and QCD < 10%). GLCM Gray-level co-occurrence matrix, GLDM Gray-level dependence matrix, GLRLM Gray-level run-length matrix, GLSZM Gray-level size zone matrix, NGTDM Neighborhood gray-tone difference matrix