Preoperative prediction of extrathyroidal extension: radiomics signature based on multimodal ultrasound to papillary thyroid carcinoma

Table 3 Comparison of radiomics values of ETE and non-ETE groups in the training set

Modality	Method	Radiomics parameters	Non-ETE (n = 151)	ETE (n = 33)	P Value
BMUS	Geometry	GeoX	(6.26 ± 0.88) × 10²	(5.88 ± 1.02) × 10²	0.028
	Grey-level co-occurrence matrix	S (5,5) InvDfMom	(1.60 ± 0.30) × 10^− 1	(1.90 ± 0.40) × 10^− 1	0.002
	Absolute gradient	GrKurtosis	2.47 ± 1.14	3.20 ± 1.01	0.001
SMI	Grey-level run length matrix	Vertl_GLevNonU	(6.85 ± 5.20) × 10²	(12.58 ± 9.07) × 10²	0.001
	Wavelets transform	WavEnHH_s5	(7.63 ± 3.18) × 10¹	(5.78 ± 1.89) × 10¹	< 0.001
	Wavelets transform	WavEnHL_s6	(1.11 ± 0.6) × 10²	(1.36 ± 0.77) × 10²	0.036
SWE	Grey-level co-occurrence matrix	H_S (0,1) SumOfSqs	(1.08 ± 0.70) × 10²	(1.03 ± 0.88) × 10²	0.008
	Autoregressive model	Teta1	(8.8 ± 0.30) × 10^− 1	(9.0 ± 0.30) × 10^− 1	< 0.001
	Grey-level run length matrix	R_Horzl_GLevNonU	(5.26 ± 4.21) × 10²	(9.64 ± 6.54) × 10²	0.001
CEUS	Grey-level run length matrix	Vertl_GLevNonU	(8.14 ± 5.87) × 10²	(15.31 ± 12.1) × 10²	0.002
	Grey-level run length matrix	Vertl_RLNonUni	(1.98 ± 1.46) × 10⁴	(3.68 ± 2.83) × 10⁴	0.002
	Wavelets transform	WavEnHL_s4	(1.04 ± 0.30) × 10²	(0.85 ± 0.20) × 10²	< 0.001
	Histogram	Perc.99%	(1.96 ± 0.15) × 10²	(1.78 ± 0.25) × 10²	< 0.001

BMUS, B-mode US; CDFI, colour Doppler flow imaging; SWE, shear-wave elastography; SMI, superb microvascular imaging; CEUS, contrast-enhanced ultrasound. S(x,y), grey level co-occurrence matrix for inter-pixel distance x along rows and y along columns; InvDfMom, inverse difference moment; Gr Kurtosis, absolute gradient kurtosis; Horzl, Horizontal; Vertl, vertical; GLevNonU, gray-level non-uniformity; WavEnHH (HL), energies of wavelet transform coefficients in frequency channels HH (HL); SumOfSqs, Sum of squares; RLNonUni, run-length non-uniformity; Perc.99%, percentile 99%

ISSN: 1471-2342