Magnetic Resonance Imaging (MRI) is a widely applied non-invasive imaging modality, with excellent soft tissue contrast and high spatial resolution. Unlike Computed Tomography (CT) scanning, MRI does not expose patients to any ionizing radiation, making it a compelling alternative. However, MRI is relatively slow compared to other imaging modalities. The total examination time can vary from 15 minutes for knee imaging to an hour or more for cardiac imaging. Remaining still for this long in a confined space is challenging for any patient, being especially difficult for children, elderly and patients under pain. Motion artifacts are not only difficult to correct, which may require a complete re-scan.

While algorithmic solutions that reduce scan time exist, like Philips’ Compressed SENSE, Artificial Intelligence is a potential game changer, promising unprecedented scan time reductions. To demonstrate this, NYU Langone Health and Facebook AI Research organized in 2019 the first fastMRI challenge, a competition where the participants were asked to reconstruct accelerated MR scans with Artificial Intelligence technology. In this talk, I will present the winning solution in the fastMRI challenge, developed by a team of researchers from Philips Research, Philips Healthcare and LUMC.

The talk is meant for a general audience and understanding of MRI is not required. I will describe a deep neural network implemented in PyTorch, so a basic understanding of deep learning can help.

Publication An Adaptive Intelligence Algorithm for Undersampled Knee MRI Reconstruction: Application to the 2019 fastMRI Challenge Authors Nicola Pezzotti, Sahar Yousefi, Mohamed S Elmahdy, Jeroen van Gemert, Christophe Schülke, Mariya Doneva, Tim Nielsen, Sergey Kastryulin, Boudewijn PF Lelieveldt, Matthias JP van Osch, Elwin de Weerdt, Marius Staring