UMRAM Seminar: “Pushing the Limits of MRI Reconstruction through Physics-Guided Deep Learning Approaches”, Dr. Ömer Burak Demirel, 10:30AM June 22 (EN)

“Pushing the Limits of MRI Reconstruction through Physics-Guided Deep Learning Approaches”

Dr. Ömer Burak Demirel
University of Minnesota
Harvard University

Date/Time: June 22, Thursday, 10:30 am
Place: Seminar Room, SC 106

Abstract: Magnetic resonance imaging (MRI) is a non-invasive diagnostic tool used in clinics to evaluate the functional properties of the human body with superior soft-tissue contrast. Scan duration is a major issue in MRI that requires trade-offs between signal-to-noise ratio (SNR), spatiotemporal resolution, and coverage leading to numerous challenges. Recently, physics-driven deep learning (PD-DL) reconstruction has gained interest in accelerated MRI. PD-DL combines MRI physics with neural network-based regularization and has shown improved image quality compared to existing methods. Yet, PD-DL has its own challenges, including limited training data availability, over-regularization at high accelerations, generalizability issues across different SNRs, and sensitivity to noise. In this talk, novel reconstruction methods will be introduced to address these challenges for state-of-the-art PD-DL reconstruction. Specific developments include: 1) A signal-intensity informed characterization of the MR encoding operator to improve the generalizability of PD-DL across different SNRs, 2) A subject-specific self-supervised PD-DL reconstruction that exploits spatiotemporal correlations by using a 3D convolutional neural network for free-breathing acquisitions, 3) A new computational fMRI pipeline that performs a synergistic combination of thermal noise suppression followed by PD-DL reconstruction. The application of these methods to cardiac and brain MRI datasets shows promise for the clinical translation of PD-DL reconstruction.

About the Speaker: Dr. Demirel is a postdoctoral research fellow at the University of Minnesota and will continue this position at Harvard University, Beth Israel Deaconess Medical Center, in the upcoming fall to continue working on deep learning applications for cardiac MRI. He received his Ph.D. in May 2023 from the Department of Electrical and Computer Engineering at the University of Minnesota. He received his M.S. degree in June 2017 and his B.S. in January 2015 from the Department of Electrical and Electronics Engineering at Bilkent University. During his Ph.D. research at the University of Minnesota, he worked with Prof. Mehmet Akçakaya on physics-guided deep learning for high-resolution MRI. Dr. Demirel is the recipient of the American Heart Association (AHA) Predoctoral fellowship for his research on “Rapid High-Resolution Whole Heart Perfusion MRI for Evaluation of Coronary Artery Disease.” He also worked on Magnetic Particle Imaging with Prof. Emine Ülkü Sarıtaş during his M.S. research.17