0617204 - ÚPT 2025 RIV US eng C - Conference Paper (international conference)
Jiřík, Radovan - Hývlová, Denisa - Macíček, Ondřej - Vitouš, Jiří - Starčuk jr., Zenon
Deep-Learning in Simultaneous DCE-DSC-MRI Perfusion Analysis.
2024 IEEE International Conference on Bioinformatics and Biomedicine (BIBM). Piscataway: IEEE, 2024, s. 4933-4941. ISBN 979-8-3503-8622-6. E-ISSN 2156-1133.
[2024 IEEE International Conference on Bioinformatics and Biomedicine (BIBM). Lisbon (PT), 03.12.2024-06.12.2024]
R&D Projects: GA ČR(CZ) GA22-10953S; GA MŠMT(CZ) LM2023050
Institutional support: RVO:68081731
Keywords : Deep learning * MRI * perfusion imaging
OECD category: Medical engineering
Result website:
https://ieeexplore.ieee.org/document/10822538DOI: https://doi.org/10.1109/BIBM62325.2024.10822538

The paper aims at improved reliability of magnetic resonance perfusion imaging and estimation of an extended set of biomarkers using these techniques. Magnetic resonance perfusion imaging is an important experimental methodology with main applications in diagnostics and therapy monitoring in oncology. The main two methods are Dynamic Contrast-Enhanced (DCE) Magnetic Resonance Imaging (MRI) and Dynamic Susceptibility Contrast (DSC) MRI. We combine these two methods in a simultaneous acquisition and data processing approach. For simultaneous DCE-DSC-MRI data processing, we suggest a conventional non-linear least-squares method and a method based on convolutional neural networks. We evaluated the proposed methods on realistically simulated synthetic datasets and illustrated their performance on a real dataset. Compared to the standard approach, the methods of simultaneous DCE-DSC-MRI analysis were more reliable. The two proposed methods of simultaneous DCE-DSC-MRI analysis were comparable, with the neural-network approach being computationally far more effective.
Permanent Link: https://hdl.handle.net/11104/0364175