Two-point B1 correction for CEST MRI by fusing voxel-wise interpolation and T1W voxel-clustering.

Abstract

Purpose: As a sensitive metabolic MRI technique, CEST images are easily contaminated by $B_1$ inhomogeneity due to strong dependence on saturation $B_1$ . We aim to develop an efficient and robust two-point $B_1$ -correction method.

Methods: The proposed method only acquires CEST images under two saturation $B_1$ 's, { $B_{1,\text{high}}$ , $B_{1,\mathrm{low}}$ }, with desired $B_1$ in between. Besides, voxel-wise Z- $B_1$ interpolation (branch A), we performed another Z- $T_1$ - $B_1$ calibration (branch B), which divided image voxels into bins according to the $T_{1}w$ image and fitted a Z- $B_1$ curve for each bin. To ensure each voxel adopts a better-corrected value, we fused the images corrected from both branches, according to a mask predicted by a retrospectively trained model. For validation, glutamate CEST (GluCEST) experiments of phantom and healthy volunteers were acquired on a 5T scanner. A total of 14 $B_1$ pairs from 2.4μT to 3.6μT were evaluated, with the 7- $B_1$ -correction as gold standard.

Results: Across glutamate phantoms with three distinct layouts, branch B demonstrated reliable correction performance for 14 $B_1$ pairs, achieving a mean absolute error (MAE) of Z(3 ppm) ≤ 5% in all 42 experiments. For six healthy volunteers, branch B yielded Z(3 ppm) images that closely matched the 7- $B_1$ correction, and the MAE distributions proved robust to voxel-binning, fitting strategies, and the choice of $B_1$ pair. After fusion, all volunteers displayed better structural similarity index measure (SSIM), than the lower ones corrected by either branch.

Conclusions: By only acquiring two ${B_1}^{\text{'}}s$ , our $B_1$ -correction strategy proved comparable performance to multi- $B_1$ methods, exhibiting robustness to $B_1$ selection and slice positions.