What is quantitative MRI, and why does it matter?

This post is going to be a gentle introduction to MRI, which is a type of brain imaging and my current focus of research.  I’m planning a future post or two about my research, so I thought I’d provide a little background for people who don’t know much, if anything, about MRI.  I am by training a statistician by way of industrial engineering, so I am by no means a neurologist, radiologist, physicist or other type of “-ist”, meaning this is going to be pretty basic.  This particular post will be about something called relaxation times and quantitative MRI scans.

I got interested in neuroimaging about a year ago, thanks to Elizabeth Sweeney, who mentored me through the process of getting up to speed.  It’s a fascinating area for biostatistics, as it combines big data, nice visualizations, and interesting and impactful biological questions.  As imaging technologies like MRI get more sophisticated, scientists are really just starting to understand the details about how the brain works.  And statisticians are just starting to get involved in this learning process, so there are a lot of opportunities to look at existing questions and problems from a new, more statistical perspective.

For about the past year, I have been working on a project having to do with quantitative MRI with Elizabeth, Taki Shinohara, and people at the NINDS.  I’ll talk more about that project in a future post, but for now here’s a basic introduction to the topic.

Imagine yourself lying flat in an MRI scanner.   As the technicians turn the MRI machine on, it creates a strong magnetic field of 3-7 Tesla.  Under that magnetic field, the protons in your brain, which were previously pointing in every direction, will all align in one direction, parallel to the magnetic field.  Then, the technicians will turn the magnetic field off, and your brain’s protons will go back to pointing their original directions.  MRI scanners can measure the “T1 relaxation times”, or the time it takes for your protons to “relax” back to their original positions.  At least, they can try to.  More on that in a bit.

But first, why measure the T1 relaxation time?  The interest in relaxation times began in 1971, when it was discovered that tumors have higher relaxation times than regular tissue.  Since then, studies have established that people with certain disorders, including multiple sclerosis (MS), tend to have widespread increases in T1 relaxation times, indicating neurological degradation, including loss of axons and myelin in the brain.  And while some affected tissue, like tumors and lesions, are easy to see on a regular, unitless MRI scan, other tissue with elevated T1 doesn’t look visibly different on MRI than healthy tissue.  That’s where quantitative images that measure T1 (“T1 maps”) come in.  Though affected tissue may not look different to the naked eye on the scan, the measured T1 tells the story.  If we can measure T1 in normal-appearing gray and white matter, we can better understand the widespread neuro-degenerative impacts of diseases like MS.  Through longitudinal studies of T1, we can better understand long-term disease burden and hopefully design better treatments and relapse detection methods.

So if T1 maps are so great, what’s the problem?  There are several major issues with T1 maps, compared with the more common clinical MRI scans, such as T1-weighted, which offer a contrast image but lack interpretable units.  Here’s an example of a T1 map (A) and a T1-weighted image (B).

• As is clearly seen, the T1 map has high levels of noise and poor signal to noise ratio compared with the T1-weighted image.  This introduces high levels of variability into the estimation of T1 levels and makes detecting subtle differences in T1 much more difficult.
• T1 maps are more difficult to obtain, since the MRI machine must be very carefully calibrated.
• They take longer to acquire, which can be problematic for some patient groups.
• T1 maps are not included in most standard protocols.
• Historically T1 maps have not been routinely acquired, which means they are missing from many long-term, longitudinal studies of multiple sclerosis and other neurodegenerative diseases.

This is where my research comes in.  Given a standard protocol, consisting of 4 clinical MRI scans, we want to predict the T1 map.  If we can do this with a reasonable degree of accuracy, we can overcome the problems above, as our estimated T1 maps will:

• Have better signal to noise ratio, similar to the contrast images used in the T1 map prediction model.
• Be easy to obtain using only the standard protocol of T1-weighted, T2-weighted, FLAIR and PD-weighted volumes.
• Not require any additional scan time.
• Be retroactively available for longitudinal and cross-sectional imaging studies of MS and other neurodegenerative diseases.

Stay tuned for a future post on how we are trying to do this.