. 2021 Jul 22;11:702055. doi: 10.3389/fonc.2021.702055

Table 3.

The definitions for the features selected for the predictive model construction.

Feature name	Feature definition and meaning
	Formula
wavelet-HHL_firstorder_RootMeanSquared_CT	$F_{s t a t, r m s} = \sqrt{\frac{Σ_{k = 1}^{N_{v}} (X_{g l, k}^{2})}{N_{v}}}$
	where N_v represents the number of voxels,X_d represents the set of intensities of the N_v voxels included in the ROI intensity mask, which could be denoted as $X_{g l} = {X_{g l, 1}, X_{g l, 2} \dots X_{g l, N_{v}}} .$ Root mean square is the square-root of the mean of all the squared intensity values, which is a measure of the magnitude of the image values.
	Formula
wavelet-LHH_firstorder_Skewness_PET	$F_{i h, s k e w} = \frac{\frac{1}{N_{v}} Σ_{k = 1}^{N_{v}} {(X_{d, k} - μ)}^{3}}{{(\frac{1}{N_{v}} Σ_{k = 1}^{N_{v}} N_{v} {(X_{d, k} - μ)}^{2})}^{\frac{3}{2}}}$
	where N_v represents the number of voxels, X_d is the set of discretized intensities of the voxels in the ROI intensity mask, which could be denoted as $X_{d} = {X_{d, 1}, X_{d, 2}, \dots X_{d, N_{v}}}$ and μ is the average discretized intensity of N_v voxels in the ROI intensity mask. Skewness is a measure of the asymmetry of the distribution of values about its mean by applying a wavelet filter, and its value could be positive or negative depending on the position that its tail is elongated and the position that the mass of the distribution is concentrated.