Table 3.
Selected features and the coefficients of features in final radiomics model.
| Features | Coefficients in model |
|---|---|
| BPGT vs. MPGT | |
| CE-T1WI_wavelet-HLL_ GLCM_ autocorrelation | 2.233 |
| ADC_ wavelet-LHL_ GLCM_ cluster shade | 1.698 |
| CE-T1WI_ wavelet-HLL_ NGTDM_ complexity | -0.355 |
| FS-T2WI_ wavelet-HLL_ GLDM_ small dependence emphasis | -0.422 |
| ADC_ original_ shape_ sphericity | -0.499 |
| ADC_ wavelet-LHH_ GLCM_ mcc | -1.488 |
| PA vs. MPGT | |
| ADC_ wavelet-LHL_ GLCM_ cluster shade | 2.890 |
| FS-T2WI_ wavelet-HLH_ GLSZM_ size zone non-uniformity normalized | 1.620 |
| CE-T1WI_ wavelet-HLL_ GLCM_ autocorrelation | 1.388 |
| ADC_ wavelet-LLH_ GLCM_ correlation | -0.135 |
| ADC_ wavelet-LHL_ first-order_ skewness | -0.342 |
| CE-T1WI_ wavelet-HLL_ GLSZM_ zone entropy | -0.881 |
| ADC_ original_ shape_ sphericity | -3.566 |
| WT vs. MPGT | |
| ADC_ wavelet-HHH_ GLSZM_ zone variance | 1.512 |
| CE-T1WI_ wavelet-HLH_ GLRLM_ run-variance | 1.033 |
| ADC_ wavelet-HHH_ GLSZM_ large area emphasis | 0.223 |
| FS-T2WI_ wavelet-HHH_ GLSZM_ gray level non-uniformity | -1.020 |
| PA vs. WT | |
| ADC_ wavelet-LHL_ first-order_ median | 3.509 |
| CE-T1WI_ wavelet-LLH_ first-order_ kurtosis | 1.102 |
| FS-T2WI_ wavelet-LLL_ first-order_ skewness | 0.920 |
| ADC_ wavelet-HLH_ GLCM_ correlation | 0.535 |
| CE-T1WI_ wavelet-LLH_ GLCM_ idn | 0.340 |
| ADC_ original_ first-order_ 10percentile | -0.531 |
| FS-T2WI_ wavelet-HHL_ GLCM_ small dependence high gray level emphasis | -1.413 |
| ADC_ wavelet HHH_ GLSZM_ size zone non-uniformity normalized | -1.504 |
GLCM, gray-level co-occurrence matrix; GLDM, gray-level dependence matrix; GLRLM, gray-level run length matrix; GLSZM, gray-level size zone matrix; Vs, versus; PA, pleomorphic adenomas; WT, Warthin tumor; BPGT, benign parotid gland tumor, MPGT, malignant parotid gland tumor; FS-T2WI, fat-saturated T2-weighted image, CE-T1WI: contrast-enhanced T1-weighted image.