Fingerprinting with MRI

A new approach to magnetic resonance imaging data acquisition and analysis promises to improve sensitivity and facilitate quantitative measurements.

Magnetic resonance fingerprinting maps (bottom) are tolerant to errors caused by a human subject moving his head (top). Reprinted from Nature.

Despite the wide use of magnetic resonance techniques in basic research and in the clinic, methodological improvements that sharpen these tools are always useful.

Mark Griswold and his team at Case Western Reserve University have been pursuing ways to improve magnetic resonance imaging (MRI) technology, in particular for clinical translation. “One of the biggest problems in MRI has been the fact that it's completely qualitative in most cases,” he notes. “Most of the time when we look at an image, we see that it's bright on one side and dark on another, and that tells us something about the disease that's there, but it's always a relative measure.” Past attempts to solve this problem have substantially increased data acquisition time or lowered measurement sensitivity and have therefore not been practical for use in a patient.

Griswold's team's new approach to MRI data acquisition and analysis promises to change this. Their method, dubbed 'magnetic resonance fingerprinting', or MRF, provides a conceptually new way of acquiring and processing magnetic resonance data.

In an MRI scan, magnetic resonance parameters—which can include longitudinal (T₁) and transverse (T₂) relaxation times, off-resonance frequency, spin density, diffusion and magnetization transfer—are serially collected. These parameters reflect the microenvironment of proton spins in water molecules in a tissue or material. The images created are qualitative, with the contrast typically weighted to just one type of parameter.

Instead the serial scanning for various parameters, in MRF the parameters are continuously varied during data acquisition such that the signals arising from the sample constitute a unique 'fingerprint'. The heart of the method is a pattern recognition algorithm that matches these fingerprints to a dictionary of predicted signal patterns for the tissue or material under investigation. One needs to know only the range of signal patterns to expect to predict the dictionary entries, notes Griswold. “In the human body, we know the range of T₁s, we know the range of T₂s, we know the range of off-resonances [and so on],” he explains.

In this way, MRF allows the researchers to assign hard numbers to the signals they detect rather than just generating qualitative maps. Such a feature should be useful not only in the clinic but for all research using magnetic resonance techniques, including nuclear magnetic resonance spectroscopy. The method also provides substantially higher sensitivity and precision than any previous method. “That and always being able to put quantitative numbers on it should really change the way we do things,” says Griswold.

Because MRF is based on pattern recognition, it is much more tolerant than conventional MRI to measurement errors, such as those arising when a patient getting a brain scan moves his or her head slightly. MRF could therefore allow clinicians and researchers to obtain higher-quality results from current-generation scanners or even to obtain satisfactory results using cheaper scanners.

An interesting potential application of MRF is the detection of early cancerous cells in tissue. “We know that cancer cells are different at the molecular level; they're different at the microscopic level,” explains Griswold. Such differences would be reflected as unique cancer cell fingerprints by MRF, potentially helping to improve diagnoses. Although his lab is largely focused on medical applications, he is also working with collaborators in basic research fields to implement MRF. “Hopefully we can get higher and higher sensitivity across the board if we do this right,” he says.