Algorithm 1: Learning the 4th-order LBPs.

(1)For the training COVID-19 CT image, compute the vector of empirical probabilities (frequencies), $F_r=[f_r\left(h\right):h=1,\dots ,96]$ of the above signal interactions. (2)Compute the frequency $F_{ref:r}=[f_{ref:r}\left(h\right):h=1,\dots ,96]$ of the same signal interactions from the synthesized image, sampled from the learned second-order MGRF, acting as a base field. (3)Compute the potentials, $V_r\left(h\right)={\lambda }_h$×$(f_r\left(h\right)-f_{core:r}\left(h\right))$, ${\lambda }_h=1/(f_{core:r}\left(h\right)$× $(1-f_{core:r}\left(h\right)))$. (4)Compute total Gibbs energy of image for candidate radius $r=1:1:10$, choose r with the largest Gibbs energy.