Restoration of Whole-Body MRI for Intensity Nonuniformities with Discontinuities

27 Jun 2025 (modified: 28 Jun 2025)Greeks in AI 2025 Symposium SubmissionEveryoneRevisionsBibTeXCC BY 4.0
Keywords: Whole-body MRI, intensity homogenization, cooccurrence statistics, Bayesian coring, anisotropic smoothing
TL;DR: Whole-body MRI uses multiple coils that produce images with spatially discontinuous intensity (shadows). We remove piecewise-smooth shadowing by deconvolving their effects from image statistics, back-projecting, and smoothing anisotropically.
Abstract: MRI assumes a uniform Radio-Frequency (RF) field. However, in Whole-Body (WB) MRI this is violated since the smooth coils’ sensitivities differ, which results in discontinuities of the intensity nonuniformities at the junctions between them as well as in overall nonuniformity differences. A method is presented for the joint intensity homogenization of two WB anatomic images. The effect of the spatial intensity nonuniformity on the co-occurrence statistics of the two images is modeled with non-stationary Point Spread Functions (PSFs) and is deconvolved. These statistics provide with Bayesian coring estimates of the posterior Bayesian expectations of the nonuniformity fields. These are processed with anisotropic filtering that enables piecewise smooth restorations. The implementation is iterative. The method has been validated with MRI datasets of forty-nine cancer patients compared to isotropic filtering.
Submission Number: 174
Loading