The value of magnetic resonance fingerprinting (MRF) can be seen in scans of a patient with right temporal lobe epilepsy. T1w and T2w images from the clinical scans (top); T1 and T2 maps from the 3D MRF scan (bottom). The potential epilepsy pathology (right panels) identified by MRF were not seen in the conventional MR scans.Credit: Dan Ma, Case Western Reserve University
Siegfried Trattnig, a radiologist at Medical University of Vienna, recalls feeling “completely thrilled” when he first encountered a new, quantifiable magnetic resonance (MR) technique at a conference in 2013. Called MR Fingerprinting, the single-sequence acquisition-and-reconstruction technology promised to more accurately diagnose diseases and improve patient care.
“I realized this was a promising tool for standardization in MR,” Trattnig says, “but what really fascinated me was that MR Fingerprinting is a fundamentally new approach to how we do MR.”
Standard MR images are tinted bright and dark, which radiologists interpret in order to diagnose disease or injury. But the degrees of shading in these traditional images is qualitative; contrast isn’t standardized. While conventional MR can create quantitative maps, the process is painstaking, taking well over an hour, and easily undermined by patient movement.
MR Fingerprinting, on the other hand, is not designed to immediately produce detailed images. Instead, it runs a multiparametric sequence over four or five minutes, where repetition time, flip angle and radiofrequency pulse phases are pseudo-randomly varied. Each tissue type responds to this sequence differently, displaying a unique signal evolution or ‘fingerprint’. Tissue fingerprints are independent of the scanner, coil or field strength used.
Then, just as forensic investigators run crime scene fingerprints through a database in the hope of finding a matched suspect, an algorithm compares MR fingerprints to entries in a dictionary.
Dictionary entries are precalculated, using the fundamental MR Bloch equations, and provide information about T1 and T2 relaxation times. These data are used to generate quantified T1 and T2 images. Eventually, MR Fingerprinting could produce a broad suite of images, including diffusion and flow maps, from a single sequence.
Using these quantitative images, tissue type and pathology can be assessed on a voxel basis, reducing reliance on human image interpretation for diagnosis. This approach may help identify conditions with little or no healthy comparison tissue, as with fibrosis. And if there are multiple tissue types in a voxel, it’s possible for MR Fingerprinting to calculate proportions of each1.
Flow chart of the MRF framework. Top panel shows four examples of dictionary entries representing four main tissues: cerebrospinal fluid — CSF (T1 = 5000 ms, T2 = 500 ms); fat (T1 = 400 ms, T2 = 53 ms); white matter (T1 = 850 ms, T2 = 50 ms); and gray matter (T1 = 1300 ms, T2 = 85 ms). A voxel fingerprint is matched to the closest dictionary entry, providing tissue features of that voxel (middle). Bottom panel shows the intensity variation of a voxel across the undersampled images — the fingerprint.Credit: Vikas Gulani, Case Western Reserve University
From pioneering project to patients in practice
MR Fingerprinting was devised by a team led by Case Western Reserve University engineer, Mark Griswold, who published a proof-of-concept2. Shortly after, in 2013, Siemens Healthineers formed an exclusive, multi-year partnership with Griswold’s Case Western group and the University Hospitals in Cleveland, in the United States, to develop the technology commercially, says Heiko Meyer, Director MR Neurology and Orthopedics at Siemens Healthineers.
The most recent stage of the collaboration involved gathering feedback. “Over the last three years, we’ve also had clinical partners in Austria, China, Germany and Japan,” Meyer says. “We’re very much interested in collaborating with more partners around the world to further refine the product.”
Trattnig’s team was one of the first to use MR Fingerprinting in a clinical setting. Their studies found T1 and T2 relaxation time differences between low- and high-grade gliomas, and between acute and chronic brain lesions in multiple sclerosis. And while MR Fingerprinting’s initial focus was the brain, it has also been applied to other parts of the body, including the abdomen3, breast4 and heart5. The next major target, Meyer says, is the prostate6.
From the patient’s perspective, MR Fingerprinting’s quantitative output means diseases should be more precisely diagnosed and monitored. Its ability to diagnose lesions and tumours could eliminate the need for many biopsies. Another benefit is that the clanging that accompanies conventional MR can, in MR Fingerprinting, be ‘tuned’ to instead play music7.
Standardized data generated by the technique might also train another assistant for radiologists: deep learning algorithms that spot early signs of disease in MR images. “These networks could tease out very subtle changes that even a radiologist would find hard to see,” Meyer says.
Trattnig is convinced that MR Fingerprinting will change the way he performs MR examinations. “It is the best way to personalize medicine, and I think that’s really exciting.”