CT diffusion trajectory

A real axial CT is corrupted with one fixed Gaussian field. The reverse view retraces that path to show what a learned denoiser must estimate.

Forward noising q(x_t | x_0)

real CT · x_0

x_0 x_T

signal variance alpha_bar_t 1.00

noise variance 1 - alpha_bar_t 0.00

x_t = sqrt(alpha_bar_t) x_0
+ sqrt(1 - alpha_bar_t) epsilon

Educational illustration: the forward states follow the stated corruption process. The reverse playback is an idealized trajectory, not output from a trained model and not synthetic clinical data.