MeMGB-Diff: Memory-Efficient Multivariate Gaussian Bias Diffusion Model for 3D bias field correction

Xingyu Qiu, Dong Liang, Gongning Luo, Xiangyu Li, Wei Wang, Kuanquan Wang, Shuo Li

Abstract: Highlights•The first 3D diffusion model is used to correct the bias field with high accuracy.•Bias fields replace Gaussian noise in diffusion models, enabling faster correction.•Novel bias trend loss learns intensity trends, which mainly causes MRI inhomogeneity.•Synthetic and clinical MRI assessments show SOTA performance.