A CVAE-based generative model for generalized B<sub>1</sub> inhomogeneity corrected chemical exchange saturation transfer MRI at 5 T.
Authors
Affiliations (3)
Affiliations (3)
- Paul C. Lauterbur Research Center for Biomedical Imaging, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, 1068 Xueyuan Boulevard, Xili, Nanshan, Shenzhen, 518055, Guangdong, China.
- Research Center for Medical AI, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, 1068 Xueyuan Boulevard, Xili, Nanshan, Shenzhen, 518055, Guangdong, China.
- Paul C. Lauterbur Research Center for Biomedical Imaging, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, 1068 Xueyuan Boulevard, Xili, Nanshan, Shenzhen, 518055, Guangdong, China; Key Laboratory of Biomedical Imaging Science and System, Chinese Academy of Sciences, 1068 Xueyuan Boulevard, Xili, Nanshan, Shenzhen, 518055, Guangdong, China; State Key Laboratory of Biomedical Imaging Science and System, 1068 Xueyuan Boulevard, Xili, Nanshan, Shenzhen, 518055, Guangdong, China. Electronic address: [email protected].
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<sub>1</sub> level. Spatial inhomogeneity of B<sub>1</sub> field would bias CEST measurement. Conventional interpolation-based B<sub>1</sub> correction method required CEST dataset acquisition under multiple B<sub>1</sub> levels, substantially prolonging scan time. The recently proposed supervised deep learning approach reconstructed B<sub>1</sub> inhomogeneity corrected CEST effect at the identical B<sub>1</sub> as of the training data, hindering its generalization to other B<sub>1</sub> levels. In this study, we proposed a Conditional Variational Autoencoder (CVAE)-based generative model to generate B<sub>1</sub> inhomogeneity corrected Z spectra from single CEST acquisition. The model was trained from pixel-wise source-target paired Z spectra under multiple B<sub>1</sub> with target B<sub>1</sub> 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<sub>1</sub> inhomogeneity corrected CEST MRI. Results showed that the generated B<sub>1</sub>-corrected Z spectra agreed well with the reference averaged from regions with subtle B<sub>1</sub> inhomogeneity. Moreover, the performance of the proposed model in correcting B<sub>1</sub> inhomogeneity in APT CEST effect, as measured by both MTR<sub>asym</sub> and [Formula: see text] at 3.5 ppm, were superior over conventional Z/contrast-B<sub>1</sub>-interpolation and other deep learning methods, especially when target B<sub>1</sub> were not included in sampling or training dataset. In summary, the proposed model allows generalized B<sub>1</sub> inhomogeneity correction, benefiting quantitative CEST MRI in clinical routines.