Table 2.

Eleven prognostic radiomics features selected by the LASSO algorithm.

Features	Descriptions	Coefficients
First order_ Skewness	Skewness measures the asymmetry of the distribution of values about the Mean value.	-2.7 × 10-1
GLSZM_ Gray Level Variance	Measuring the variance in gray level intensities for the zones.	4.32 × 10-1
Shape _Sphericity	Measuring the roundness of the shape of the tumor region relative to a circle	1.13 × 10-1
Shape _ Surface Volume Ratio	A lower value indicates a more compact (sphere-like) shape and dependent on the volume of the ROI.	2.46 × 10-2
GLCM _ Contrast	Measuring the local intensity variation, favoring values away from the diagonal.	2.87 × 10-3
wavelet-HLL_ GLDM_ DNU	Describing the homogeneity among dependencies in the image. The value is low if the image has more similarity.	-5.1 × 10-1
wavelet-LHL_GLCM_ Autocorrelation	Describing the magnitude of the fineness and coarseness of texture.	-6.72 × 10-2
wavelet-HLH_ NGTDM_ Busyness	Describing the change from a pixel to its neighbor. The value is high if the changes of intensity between pixels and its neighborhood is rapid.	-1.91
wavelet-LLL_ NGTDM _Complexity	Describing the complexity of the image. The value is high if there are many rapid changes in gray level intensity.	4.09 × 10-3
wavelet-HLL_ GLSZM_ SAHGLE	Describing the distribution of smaller size zones with higher gray-level values.	-1.17 × 10-1
wavelet-LLH_ GLSZM _ SZNUN	Describing the variability of size zone volumes throughout the image.	-7.42 × 10-2

DNU, Dependence Non Uniformity; SAHGLE, Small Area High Gray Level Emphasis; SZNUN, Size Zone Non Uniformity Normalized.