MRI is great for imaging tissues and organs, as it does not involve any invasive procedures (such as drugs, radiation or even needles/scalpels). It allows us to quickly and safely get a good idea of what goes on under the skin. However, just as with a photo, it can become pixelated and useless, especially if you are looking at small structures. Imagine taking a holiday snapshot of a distant landmark, say, the Eiffel tower, but when you zoom in, it becomes hard to see what is tower and what is sky. Up close, the image is hard to interpret, and determining the exact size of the structure in the image may become impossible. In a pixelated image of the Eiffel tower, one pixel may contain both tower and sky, and you can’t tell where the exact line between the two goes. This is also the case with MRI.
The MRI image is divided into voxels (same as pixels, just three-dimensional – think of it as a set of tiny cubes making up the image). The quality of your image depends on the number of voxels and the signal from these. Too few voxels, not enough information. It would be like having just a few pixels covering the spire of the Eiffel tower – it won’t look good. One voxel covering a big chunk of the brain of a human participant won’t be very useful as there simply is not enough resolution to determine what’s what.
You can also have enough voxels but too little signal. Too little signal means not enough information. It would be like taking the photo without any light. In photography, light creates the image. If you have plenty of light, you can typically get more detail from each part of your image, and so your resolution can be better. Not enough signal means you need to keep your shutter open for longer, to let more light in. You can do that, but you probably need a tripod, and any movement (birds, people, wind, clouds, a bus rumbling past) will affect your picture and make it blurry. Same with MRI. You can increase signal by increasing scan time, but that’s not always possible and means you have to keep people still and in the scanner for longer.
Typically, the stronger your field strength, the more signal (light) you can get from your voxels. More signal means you can reduce voxel size as you will be getting more detail from each part of your scan. This typically results in better resolution. Or, you can get the same resolution, just faster (i.e. keep the shutter open for a shorter period). This can be useful if what you’re wanting to scan moves. For some tissues, it is invaluable to be able to get quick images. Imagine getting an image of a beating heart, for example. You need a quick scan sequence, and you typically need to do it several times over to get a nice, detailed image. That is the equivalent of taking lots of quick shots of the Eiffel tower (swaying in the wind, perhaps), and piecing them together to get all the fine details. This process of getting many snapshots of the same thing and taking the average to get a good image is very common in MRI, and it helps with a third issue: noise.
Everything you see in an MRI image is either signal, or it is noise. We call the relationship between signal and noise a signal-to-noise ratio (SNR), and we want it to be as high as possible (more signal, less noise). Noise in MRI images is usually caused by particles with an electrical charge moving around slightly in the human body, or by electrical resistance in the MRI machine itself. Together, these cause variations in signal intensity. Again, in our photo of the Eiffel tower, this is much the same: small visual distortions that gives it a grainy quality.
When we have smaller voxels, we typically get less signal and more noise per voxel (a low SNR). The way around it is to increase the number of averages that we run – i.e. take more snapshots, get a better image. Unfortunately, this takes time. Usually, then, we end up with a compromise of how good a resolution we want and how long we want the scanning to take, and this is greatly dependent on the kind of signal we can get.
The good news is that signal is, as mentioned earlier, grossly dependent on the field strength of our magnet. And we have some strong magnets available to us. For a human scan with a field strength of 3 Tesla (T), a resolution of 1mm x 1mm is easily obtained in about 5 minutes. That is fine for a structural scan of, say, a human brain. But what if you want to scan something smaller? Humans move, no matter how hard they try not to, even if it is only to draw breath. Higher resolutions means you’ll pick up on these small movements. A tiny motion can shift a small voxel completely out of place, while a bigger voxel would be less affected. In short, smaller makes things more difficult.
So, how small is too small?
For higher-field machines in humans, such as 7T, you can go small. 0.5mm x 0.5mm for 7T is easily doable, and some post mortem scans has gone down to 0.14mm x 0.14mm . For our 9.4T scanner, we have scanned with resolutions of 0.03mm x 0.03mm (post mortem), and resolutions at 0.06mm x 0.06mm should be possible for scans that are not post mortem. That means with 9.4T, we can in theory image structures as small as 0.25mm across with some clarity (remember, you need to have at least a few voxels of the thing you want to image to properly see the edges and detail of it). Our scanner is too small for a whole human, but there exists 9.4T MRI machines for humans. And an 11.75T magnet is underway (see the BBC news story here), which will be able to get resolutions down to 0.1mm (or possibly more) in humans. So at present, anything less than 0.25mm is probably too small for MRI.
Smaller than that, and we have to use other methods. Microbiologists and their microscopes are likely to laugh in the face of 0.25mm. Below is a picture with a 9.4T image and histology image of testicular tissue , showing how much detail both techniques afford. Histology is obviously better. Sadly, however, it still requires scalpels, and MRI does not.
 Stucht D, Danishad KA, Schulze P, Godenschweger F, Zaitsev M, Speck O. Highest Resolution In Vivo Human Brain MRI Using Prospective Motion Correction. PLoS ONE. 2015;10(7):e0133921. Link: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4520483/