A CVAE-based generative model for generalized B1 inhomogeneity corrected chemical exchange saturation transfer MRI at 5 T

Abstract

Chemical exchange saturation transfer (CEST) magnetic resonance imaging (MRI) has emerged as a powerful tool to image endogenous or exogenous macromolecules. CEST contrast highly depends on radiofrequency irradiation $B_1$ level. Spatial inhomogeneity of $B_1$ field would bias CEST measurement. Conventional interpolation-based $B_1$ correction method required CEST dataset acquisition under multiple $B_1$ levels, substantially prolonging scan time. The recently proposed supervised deep learning approach reconstructed $B_1$ inhomogeneity corrected CEST effect at the identical $B_1$ as of the training data, hindering its generalization to other $B_1$ levels. In this study, we proposed a Conditional Variational Autoencoder (CVAE)-based generative model to generate $B_1$ inhomogeneity corrected Z spectra from single CEST acquisition. The model was trained from pixel-wise source–target paired Z spectra under multiple $B_1$ with target $B_1$ as a conditional variable. Numerical simulation and healthy human brain imaging at 5 T were respectively performed to evaluate the performance of proposed model in $B_1$ inhomogeneity corrected CEST MRI. Results showed that the generated $B_1$-corrected Z spectra agreed well with the reference averaged from regions with subtle $B_1$ inhomogeneity. Moreover, the performance of the proposed model in correcting $B_1$ inhomogeneity in APT CEST effect, as measured by both $MT{R}_{asym}$ and $MT{R}_{R_{ex}}$ at 3.5 ppm, were superior over conventional Z/contrast-$B_1$-interpolation and other deep learning methods, especially when target $B_1$ were not included in sampling or training dataset. In summary, the proposed model allows generalized $B_1$ inhomogeneity correction, benefiting quantitative CEST MRI in clinical routines.