An interpretable machine learning model based on computed tomography radiomics for predicting programmed death ligand 1 expression status in gastric cancer

Table 2 Comparison of radiomics features between PD-L1 high and low expression groups in training set

Features	Radiomics feature value (Z-score normalization)
Features	PD-L1 low Expression (n = 73)	PD-L1 high Expression (n = 70)	P value
wavele_HH_glcm_Idn	0.25 ± 0.99	-0.26 ± 0.95	0.002
wavele_HL_gldm_Dependence-Entropy	0.21 ± 0.98	-0.22 ± 0.97	0.009
wavele_HL_glszm_LargeAreaLowGrayLevelEmphasis	-0.18 ± 0.70	0.19 ± 1.21	0.005
wavele_LL_firstorder_Kurtosis	0.25 ± 1.24	-0.26 ± 0.56	0.009
wavele_LL_firstorder_Minimum	-0.29 ± 1.04	0.30 ± 0.87	< 0.001
wavele_LL_glcm_Idn	0.28 ± 0.96	-0.29 ± 0.97	< 0.001
wavele_LL_glcm_JointEntropy	0.25 ± 0.95	-0.27 ± 0.99	0.002
wavele_LL_glcm_Maximum-Probability	-0.22 ± 0.77	0.23 ± 1.16	0.015
wavele_LL_ngtdm_Busyness	-0.25 ± 0.73	0.26 ± 1.17	0.009

Note: Data are presented as mean ± standard deviation, and compared using an independent t-test
Abbreviations: CPS, combined positive score; glcm, gray level cooccurrence matrix; glszm, gray level size zone matrix; ngtdm, neighboring gray tone difference matrix

ISSN: 1470-7330