Project ID BE-MI2023_07


Co Supervisor 1A School of Biomedical Engineering & Imaging SciencesWebsite

Co Supervisor 1B School of Biomedical Engineering & Imaging SciencesWebsite

Real-time deep-learned patient-specific safety monitoring for ultra-high field MRI

Parallel-transmit (pTx) ultra-high field MRI using multiple radiofrequency (RF) transmitters demonstrates impressive advantages for high-resolution anatomic and functional imaging of the human body. By adjusting the RF phases and amplitude of individual transmitters pTx allows control of transmit uniformity and efficiency. However, this flexibility can result in an increased range of potential excessive RF power absorption levels in the body, leading to unwanted tissue heating. Currently, no direct measurement method is available for excessive tissue heating monitoring for MRI scans, and current methods are mainly based on population-base estimation of local excessive tissue heating with safety margins for subject-specific variations using computational electromagnetic (EM) simulations. In this project, we aim to achieve patient specific real-time tissue heating monitoring during MRI scans to enable safe and efficient imaging. The project involves the following investigations:

State-of-the-art methods for segmentation (e.g. U-Nets) will be trained for accurate detection of regions of interest from 3D images of detailed tissues

Use of fast, localizer images to detect the critical regions of interest by implementing generative models (e.g. GANs) to generate water/fat images from scout images for high quality segmentations, enabling fast, accurate and patient-specific safe ultra-high field MR scanning

Patient-specific monitoring by simplified human abdomen models with a limited tissue types in simulations instead of whole-body full tissue models (Figure 1)

Automated segmentation of 10 second localizer images during MR scans will provide patient-specific specifications for fast, real-time safety monitoring.

One representative publication from each co-supervisor:

Ipek Ö, Raaijmakers AJ, Lagendijk JJ, Luijten PR, van den Berg CAT. Intersubject local SAR variation for 7T prostate MR imaging with an eight-channel single-side dipole antenna. Magnetic Resonance in medicine 2014; 71:1559-1567.

Mehranian A, Reader AJ. Model-Based Deep Learning PET Image Reconstruction Using Forward-Backward Splitting Expectation-Maximization. IEEE Trans Radiat Plasma Med Sci. 2020 Jun 23;5(1):54-64. doi: 10.1109/TRPMS.2020.3004408. PMID: 34056150; PMCID: PMC7610859.