New Magnetic Resonance Fingerprinting Technology Makes Process Faster, Cheaper, More Accurate

CDS and Langone researchers receive distinction from international MRI organization

Magnetic resonance imaging (MRI) produces images of internal organs by sending a series of magnetization pulses through a body placed inside a strong magnetic field. In traditional MRI the rich informational content of the magnetic resonance pattern of each tissue is reduced to an image of signal intensities (also called contrasts) which is visually examined by a radiologist to identify the relevant tissues. This qualitative process of identifying tissues is inherently subjective and further complicated by the fact that MRI scanners produced by different manufacturers often have different field intensities. Consequently, images of the same tissue from different scanners may look different and may be difficult to compare.

A new modern perspective has emerged: In 2013, a new MRI technique called magnetic resonance fingerprinting (MRF) was developed to identify tissues quantitatively. The technique is called fingerprinting because the magnetic resonance pattern of each tissue is unique. This means that a computer system, rather than a visual analysis of image contrasts, can identify the relevant human tissue by examining its magnetic resonance pattern. However, magnetic field inhomogeneities of scanners affect these patterns. These inhomogeneities must be addressed for a computer system to identify tissues consistently across scanners.

So Vlad Kobzar, CDS PhD candidate, Carlos Fernandez-Granda, Assistant Professor of Mathematics and Data Science at CDS and Courant Institute, and Jakob Assländer, Assistant Professor at NYU Langone Department of Radiology, designed a pulse sequence that allows tissue identification in MRF to be independent of magnetic field inhomogeneities. This research combines mathematics and physics (differential equations governing magnetic resonance) with statistics and optimization (maximizing the quality of identification over all possible pulse sequences). It provides a new roadmap for consistently identifying tissues across a broad array of hardware.

Tissue identification that does not depend on hardware characteristics would make it easier to compare MRI scans from different scanners. The new method would also accelerate the procedure, reduce its cost, and improve its accuracy. Ultimately, these advantages all benefit patients.

The authors had their paper accepted for publication by the International Society for Magnetic Resonance in Medicine (ISMRM), and the paper received the distinction of being selected for an oral presentation at the upcoming ISMRM Annual Meeting. The authors were supported by the Moore-Sloan Data Science Environment at NYU, the NSF, and the NIH.

By Paul Oliver