Back
Linear projection-based CEST reconstruction: the simplest explainable AI
{Evaluation of multi-parametric in vivo CEST MRI often requires complex computational processing for both field inhomogeneity correction and contrast generation. In this work, linear regression was used to obtain coefficient vectors that directly map uncorrected 7T spectra to corrected Lorentzian target parameters by simple linear projection. The method generalizes from healthy subject training data to unseen test data of both healthy subjects and tumor patients. The linear projection approach thus integrates correction of both B0 and B1 inhomogeneity as well as contrast generation in a single fast and interpretable computation step.}
@misc{item_3319963, title = {{Linear projection-based CEST reconstruction: the simplest explainable AI}}, booktitle = {{2021 ISMRM \& SMRT Annual Meeting \& Exhibition (ISMRM 2021)}}, abstract = {{Evaluation of multi-parametric in vivo CEST MRI often requires complex computational processing for both field inhomogeneity correction and contrast generation. In this work, linear regression was used to obtain coefficient vectors that directly map uncorrected 7T spectra to corrected Lorentzian target parameters by simple linear projection. The method generalizes from healthy subject training data to unseen test data of both healthy subjects and tumor patients. The linear projection approach thus integrates correction of both B0 and B1 inhomogeneity as well as contrast generation in a single fast and interpretable computation step.}}, year = {2021}, slug = {item_3319963}, author = {Glang, F and Fabian, M and German, A and Khakzar, K and Mennecke, A and Laun, F and Kasper, B and Schmidt, M and Doerfler, A and Scheffler, K and Zaiss, M} }