Learning-based quantification of magnetization transfer contrast and chemical exchange saturation transfer using MR fingerprinting

Abstract

Magnetization transfer contrast (MTC) and chemical exchange saturation transfer (CEST) are molecular MRI techniques that provide functional information of semisolid macromolecules and mobile solute molecules such as metabolites, proteins, and peptides. To overcome the low sensitivity due to their low population, the repeated RF saturation of target protons results in a decrease of water signal through magnetization transfer, thereby allowing the indirect assessment with improved sensitivity. However, the contrast based on ratio of the signal reduction is highly dependent on scan parameters and water tissue-relaxation effects. We introduced magnetic resonance fingerprinting (MRF) technique that estimates multiple tissue parameter simultaneously for tissue quantification. Unique signal profiles, so called fingerprints, were obtained by pseudo-randomly varying the imaging parameters for each scan and used to estimate multiple tissue parameters. The quantification accuracy highly depends on the design of the MRF acquisition schedule. Thus, we proposed a learning-based optimization framework to accelerate data acquisition and improve quantification accuracy. The acquisition parameters were updated towards minimizing the loss function that directly calculated tissue quantification errors through neural network. Therefore, the proposed method improved parametric reconstruction quality compared to conventional optimization techniques that used indirect objective functions. In addition, we proposed B$_0$ and B$_1$ inhomogeneity correction method to compensate the inhomogeneity-induced artifacts and errors. Especially, since the B$_1$ inhomogeneity changes RF saturation power, B$_1$ errors in MTC- and CEST-MRF containing repeated RF irradiation would impair the tissue quantification. The proposed method calibrated the scan parameters using the measured B$_0$ and B$_1$ maps in dynamic scan-wise manner, which allowed accurate tissue parameter mapping. We further developed the correction technique to correct the B$_0$ errors without the additional B$_0$ map, which even reduced the scan time. We demonstrated that the proposed approach could correctly estimate MTC and CEST tissue parameters, even with severe B$_0$ and B$_1$ inhomogeneities.