Dr. Olivier Scheidegger, Inselspital Bern

Abstract (Lay summary see below)

Magnetic resonance (MR) imaging of peripheral nerves, termed MR neurography (MRN), is increasingly gaining attention for clinical diagnostic evaluation and treatment monitoring of peripheral nerve diseases. It allows morphologic investigation of proximal and deeply situated nerve segments not accessible by other diagnostic techniques. MR neurography is now clinically used as a complementary tool to electrodiagnostic studies, and allows quantification of biophysical nerve tissue properties based on morphology and signal behaviour. Previously, we have developed a fully automatic nerve segmentation technique for clinical MRN images using convolutional neural networks (artificial intelligence, deep learning) and proposed a concept of quantitative analysis of nerve morphology and lesion type. Using even higher resolution MRN, qualitative analysis of nerve substructures such as nerve fascicle bundles becomes now available. However, these MRN sequences are 2-D sequences, require long acquisition times (1-2 hours for an entire extremity), and are prone to motion artifacts, limiting their extended use despite their potential of shedding new insights into nerve pathology. The aim of this project is to develop a completely new MRN sequence termed ”rapid exploratory imaging (REI)” for motion-insensitive, 3-D high-resolution imaging (down to 200-250 micrometers) that can cover a whole extremity from the nerve roots to the most distal parts within a clinical feasible scan time of around 20 minutes. The REI sequence development will follow a single slab 3-D turbo spin echo sequence with variable excitation pulses to restrain specific-absorption-rate issues. State-ofthe-art fat saturation schemes shall be applied in the REI sequence to improve image contrast. To achieve our aims of high resolution and fast acquisition times, non-Cartesian acquisition schemes combined with parallel imaging and compressed sensing reconstruction algorithms will be used. Complex iterative algorithms based on non-uniform fast Fourier transform will be employed for MR signal reconstruction, adopting variable artificial intelligence for near real-time results. Quantitative morphometric analysis of peripheral nerves will be performed using an adoption of our previous work using fully-convolutional neural networks (AI) in a prospective study of clinically and neurophysiologically defined patient cohort. This project is a straightforward corollary of our previous work and builds on a fruitful collaboration with biomedical engineers and physicists at the University of Bern and the Institute of Myology in Paris.

Lay summary

“Rapid Exploratory Imaging (REI)” - Eine neue hochaufgelöste MR Neurographie Technik für die Bildgebung peripherer Nerven

Die Diagnostik von Nervenschädigungen beruht auf der ärztlichen Untersuchung sowie neurophysiologischer Messungen der Nerven und Muskeln an Armen und Beinen. Nerven nahe der Wirbelsäule lassen sich so nur schlecht untersuchen. In den letzten Jahren hat sich als zusätzliche Abklärungsmethode die hochauflösende MR Neurographie etabliert, welche eine bildgebende Darstellung auch kleiner Strukturen innerhalb der Nerven ermöglicht. Leider ist diese Technik sehr zeitintensiv und nur auf kleine Nervenabschnitte beschränkt. Ziel dieses Projektes ist es, eine neue MR Pulssequenz zu entwickeln (“REI”), welche Nervenstrukturen in 3D in einer Auflösung von ca. 250 Mikrometer ermöglicht, und dies über grosse Nervenabschnitte und in einer kurzen Aufnahmezeit. Die Schwierigkeit besteht insbesondere darin, dass die Aufarbeitung des MR-Signals und die Bildanalyse im weiteren Verlauf selbst aufgrund der grossen Datenmenge sehr komplex ist. Mittels Einsatzes künstlicher Intelligenz soll diese Datenverarbeitung möglich sein.