PhD Candidate AI-based MR Reconstruction Methods Radiology

Magnetic Resonance Imaging (MRI) has become widespread with neuroimaging, cardiovascular, musculoskeletal, and head-and-neck imaging among the most well-known applications. However, MRI is characterized by its long scanning times. This makes MRI cost-intensive, subject to motion artefacts, and uncomfortable and stressful for patients. In addition, MRI can suffer from image artifacts arising mainly from patient or organ motion, resulting in reduced image quality, and compromised clinical utility. To increase the range of applications, diagnostic quality and outcome-per-costs ratio of MRI, we are looking at ways to reduce scan times, and simultaneously improve image quality. In this PhD project you will investigate AI-based methods for the joint reconstruction of multi-contrast images, as well as those based on magnetic resonance fingerprinting (MRF) acquisitions. Your aim is to develop AI models for sequence optimization for synthetic MRF to achieve high SNR, accurate contrast weighted images (T1W, T2W, FLAIR) with clinically relevant resolution in acceptable scan times. This project is in collaboration with Philips Electronics Nederland B.V. who, together with the LUMC participated in the fastMRI Challenge, organized by NYU and Facebook, where we won key tracks in the competition. Together, we have developed a physics-informed neural network to reconstruct knee MRI data acquired at 1.5 and 3T from 4x or 8x under-sampled k-space data, i.e. accelerating the acquisition by a factor of 4 or 8. These developments show that significant scan time reductions are achievable with AI technology. In addition, AI generates ample opportunities for advances in image quality improvements (e.g. scan robustness and image-based artefact reduction), advanced image processing applications, increased information extraction, improved patient comfort, reduction in contrast agent dose, and even possibilities for intelligent adaptive acquisition schemes.