SHAMAEI, A. HLUBOKÉ UČENÍ PRO KVANTIFIKACI JEDNOVOXELOVÝCH A MULTIDIMENZIONÁLNÍCH MR SPEKTROSKOPICKÝCH SIGNÁLŮ A JEHO SROVNÁNÍ S NELINEÁRNÍM FITOVÁNÍM METODOU NEJMENŠÍCH ČTVERCŮ [online]. Brno: Vysoké učení technické v Brně. Fakulta elektrotechniky a komunikačních technologií. 2023.
The topic of Amir Shamaei’s dissertation is estimation of metabolites concentrations using quantification of magnetic resonance spectroscopy (MRS) data. Until recently, this task has been mostly solved using conventional techniques of model fitting using standard optimization techniques. The dissertation is focused on design of new MRS quantification approaches based on deep learning (DL), as a potentially more accurate and faster approach. In the field of MRS, this direction has been explored only to a limited extent. In this dissertation, novel DL-based approaches were designed for MRS data analysis. Namely, the MRS-signal preprocessing step, including frequency and phase correction, and the MRS quantification step were investigated and appropriate new DL methods were designed. Both, supervised and self-supervised approaches were proposed. Also, new methods were designed for quantifying uncertainty in the estimated metabolite concentrations. To facilitate data sharing among research groups, the author has actively participated in design of a standard data format. The proposed methods were compared to the conventional nonlinear-least-squares fitting methods and evaluated on simulated and in-vivo data. Amir Shamaei worked on his dissertation very actively and independently, bringing up his own original solutions. During his study, he was an active member of an international consortium under the Horizon 2020 project INSPiRE-MED. Based on the feedback of this project’s senior researchers, he was one of the most productive and contributing Ph.D. students within the project. During his Ph.D. study he also spent 3 months on a secondment at the university hospital Inselspital (University of Bern, Switzerland) and one month at SME Tesla Dynamic Coils, Zaltbommel, Netherlands. His stays resulted in one IF journal paper and an additional journal paper currently shortly before submission. Amir Shamaei has published the main results of his dissertation as two first-authored and two co-authored papers in Q1 journals and several conference papers. I value Amir Shamaei as a skilled Ph.D. student and researcher and his dissertation as an excellent work. I recommend the thesis for the defense.
viz pdf
viz pdf
eVSKP id 153309