Quantitative MRI – an innovation in need of reinventing
Quantitative MRI (qMRI) is a long innovation in the making, but is finally making its way into cardiac MRI guidelines. In this blog post, Guillaume explains what qMRI is and where the hidden hurdles are. And it is not the first thing one can think of!
On average, it takes 17 years for innovations to break through into clinical practice1, but for quantitative MRI (qMRI), it has been slower. The idea behind qMRI is as old (or as new) as clinical MRI systems: it dates back from 1984. In this blog, I explore the challenges of qMRI and what can be done to address them.
True vs. arbitrary units
MRI is a powerful analytical tool in medicine; however, it is mostly used qualitatively. Doctors or researchers analyze gray level images. They rarely measure their intensity values. Images are typically T1 or T2 “weighted” (T1w or T2w), T1 and T2 being time constants linked to the tissue relaxation time. This is perfect in diagnoses, and even to measure distances, surfaces or volumes. However, each voxel’s intensity uses an arbitrary unit, and measuring it is pointless.
The point of qMRI is to measure voxels intensity values in absolute terms, expressed in true units. For example, T1 and T2 maps are expressed in msec, proton density maps in % of water protons, apparent diffusion coefficient maps in mm2/sec, cerebral blood flow maps in mL/100g/min, etc. Notice the use of “map” instead of image, hence the expression “mapping”.
qMRI: distrusted by radiologists, avoided by MRI vendors
Using an MRI system quantitatively is the exception, not the norm. qMRI is distrusted by radiologists. Years ago, a radiologist once told me that if a doctor needed a ruler to do her job, she’d ceased to be a doctor. To this day, I cannot make my mind if I believe it or not.
qMRI is also avoided by MRI vendors. One of my first jobs was to lead a team developing a brain perfusion analysis software. The parametric maps we were generating had technical names to avoid being called what they were used for. For instance, we used the “area under the curve” (mL/100g) for cerebral blood volume and “mean time of enhancement” for mean transit time (sec). We never used true units in our maps. MRI vendors are extremely cautious of not making MRI system a measurement tool for fear of falling into a more stringent regulatory category. And yet, measurement is what an MRI is used for.
The advent of synthetic MRI
Fast forward to 2015 and Swedish SyntheticMR, at the time 8 years old, founded by the brilliant inventor Marcel Warntjes, used multi-echo pulse sequences to quantify key parameters such as T1, T2 and others2. In the same year, the new concept of MR Fingerprinting (MRF) emerged. MRF exploits continuous and highly accelerated acquisitions with dictionary matching reconstructions to generate synthetic qMRI maps. As a side note, SyntheticMR, now a business worth several millions of euros, recently received CE mark for its latest version, jumping through the hoops of the EU Medical Device Regulation; its technology is now available on most MRI platforms.
qMRI, Synthetic MRI or MR Fingerprinting?
What’s the link between qMRI, synthetic MRI and MRF? From a helicopter view, it’s all the same: measuring absolute values of key physiological parameters. Zooming in, qMRI is the general approach to measure, rather than image, using an MRI. Synthetic MRI is a technique to not only measure MRI parameters, but also use them to reconstruct weighted images: T1w, T2w, etc. MRF accelerates data acquisition, and dodges the unpopular “synthetic” adjective. “Synthetic” sounds like “making stuff up” and is not aligned with a radiologist’s desire to get the best quality image one can get.
Despite its promises, and maybe due to this unfortunate name, synthetic MRI failed to become routine clinical practice. From its inception, it was intended for the brain, and is relevant in applications such as neoplasms and multiple sclerosis. However, a large study commended by GE Healthcare did not deliver on the promise. T2 flair, one of the most important sequences used in the diagnosis of brain lesions, was not 100% reliable with synthetic MRI3.
Brain and other more vital organs
Is brain the best target organ for qMRI? I don’t think so. For the past few decades, brain MRI has been a walk in the park. I am baffled by the investment to make something good better when a multitude of challenges need to address in other body organs. There are other vital organs where qMRI could play a critical role. T1 and T2 mapping are recommended in cardiac MRI, and routinely d one in the analysis of the myocardium4. Parametric MRI shows promising results in prostate or breast cancer5,6. One day, new biomarkers like the T1ρ parameter in the myocardium, might even remove the need of extraneous contrast agents aka Gadolinium7,8.
The issue with Gadolinium
Gadolinium-based contrast agents were in the spotlight a few decades ago for causing a rare and debilitating disease, nephrogenic systemic fibrosis. Recently, Gadolinium was shown to accumulate in the brain9. Gadolinium ions are highly toxic to mammals, and after being clinically used, eventually end up in waste waters, potentially polluting our only planet. Gadolinium-based contrast agents play an essential role in diagnoses, but their use should be avoided when possible. In the heart, myocardial late gadolinium enhancement (LGE), obtained a few minutes after injecting the contrast agent, is one of the most important images for the diagnosis, particularly to detect myocardial lesions. qMRI could detect and measure myocardial injuries, with recent research demonstrating that T1ρ maps might give the same information as LGE7.
The ugly truth of qMRI
As beautiful as the science of quantifying parameters is, it is challenged by a flawed assumption. Signal analysis is performed assuming that a given voxel remains in the same location with respect to the patient’s body at various measurement times. If this assumption might be valid in the brain or the prostate (playing quite similar roles in men – just my opinion), it is incorrect in more vital abdominal organs which move with respiration, cardiac contractions, or peristalsis. This makes qMRI challenging, despite its promises. Patient breathing may move tissue in and out of a scanned slice location.
To address the issue of respiratory motion, breath-holding or navigated data acquisitions sequences are common. They are not an efficient use of the MRI scanning time. The MRI scanner is idle most of the time, either because the patient must recover after breath holding, or because it is waiting for the patient’s diaphragm to return to the correct position. These pauses make MRI scanning very inefficient.
Another flawed assumption is that echoes from multi-echo pulse sequences are ideally spaced in time. Respiration and cardiac contractions impact the timing of echoes. Failing to take this into account will lead to large measurement errors10.
The future of qMRI
Despite its challenges, qMRI has a promising future. Taking patient motion into account and correcting for it is increasingly common. Precise recording of the patient’s heartbeat, respiration or other movements is now routine. Elaborated image reconstruction schemes will be required to allow, for example, myocardial T2 mapping in free-breathing11. For cardiac motion, nothing beats the detection of QRS in electrocardiograms (ECG). For breathing motion, antiquated respiratory bellows have given a nightmare to generations of technologists and researchers, and are fortunately being replaced by radar-based or camera-based techniques. Motion sensors (accelerometer, gyroscope) can also address the motion issue12.
MRI and/or third-party vendors have a role to play. They need to develop calibration devices (phantoms, etc.) and guarantee the MRI scanner’s accuracy not only at day 1, but throughout the course of time. Measuring quantitative parameters for each voxel location will provide new biomarkers and empower clinicians and researchers. Epsidy’s mission is to enable this paradigm shift through MRI compatible sensors for ECG and respiration.
Sources and references
J R Soc Med. 2011;104(12):510-520. doi:10.1258/jrsm.2011.110180
Magn Reson Med. 2008;60(2):320-329. doi:10.1002/mrm.21635
Am J Neuroradiol. 2017;38(6):1103-1110. doi:10.3174/ajnr.A5227
J Cardiovasc Magn Reson. 2017;19(1):75. doi:10.1186/s12968-017-0389-8
BMC Cancer. 2019;19(1):1244. doi:10.1186/s12885-019-6434-2
Cancer Imaging. 2020;20(1):88. doi:10.1186/s40644-020-00365-4
J Cardiovasc Magn Reson. 2021;23(1):119. doi:10.1186/s12968-021-00781-w
JACC Cardiovasc Imaging. 2021;14(10):1945-1947. doi:10.1016/j.jcmg.2021.07.020
Lancet Neurol. 2017;16(7):564-570. doi:10.1016/S1474-4422(17)30158-8
Magn Reson Mater Phys Biol Med. 2020;33(4):569-580. doi:10.1007/s10334-019-00815-6
IEEE Trans Med Imaging. 2016;35(1):197-207. doi:10.1109/TMI.2015.2463088
IEEE Trans Biomed Eng. 2017;64(1):123-133. doi:10.1109/TBME.2016.2549272
