Physics for understanding fMRI artifacts: Part One

This is the first in a series of posts in which I will attempt to provide you with the means to diagnose the manifold artifacts that plague fMRI. These artifacts can be inherent, e.g. distortion and dropout, or the consequences of a hardware issue, e.g. RF interference or gradient spiking, or may arise from your subject, e.g. cardiac pulsatility and head movement. However the artifacts arise, the aim is simple: by providing you the means to recognize what is going wrong in your experiment you may be able to discern the root cause, then remedy the problem and salvage your data.

It's a lot like plane spotting, only less fun.

The psychologists amongst you would be able to lecture me for days on category learning. Well, that's what this is. Except that before we can differentiate between one artifact and another we must first understand how EPI is designed to work in an ideal situation. (You don't need to know how planes fly to categorize them, apparently.) And to comprehend artifacts it's important for you to have a reasonable appreciation of the underlying physics, especially the concept of k-space. Here's the loose plan:

1. Review basic principles of NMR and MRI
2. Introduce the k-space representation
3. Review the EPI pulse sequence and its k-space representation
4. EPI's "classic" artifacts: ghosts, distortion and dropout
5. "Good" data: what artifacts are inherent?
6. Time-independent (or static) artifacts
7. Time-dependent (or dynamic) artifacts

It'll take many more than seven posts. This is just the order in which I'll cover the topics.

As a general rule, static artifacts are unwanted features of EPI. We can potentially reduce them but it's nigh on impossible to remove them completely. Dynamic artifacts are the real scourge. Whether they arise from the scanner or the subject, anything that varies as a function of time that isn't tied (through neurovascular coupling) to the stimulus is going to harm temporal SNR (TSNR). Of course, the only way you might be able to tackle an artifact, especially a dynamic one, is to diagnose it accurately in the first place. Hence the need to look at the entire spectrum of artifacts, from all sources and whether time-dependent or not.

Buckle up for safety!

For those of you who might have tried before to understand concepts like phase encoding and k-space but were left numb, don't worry. I intend to focus on concepts. In particular, I will try to introduce concepts that you can use in the lab, on the scanner, to help you get better data. (Yes, understanding k-space really can help you get better data! Indeed, without a reasonable understanding of k-space or the EPI pulse sequence it's almost impossible to comprehend the many sources of artifacts that can hamper, or destroy, your fMRI experiments.) So buckle up and hang on for the ride. It really won't be that bumpy, I promise.


What is NMR and how does it work?

Lucky for me (and you) there is already a considerable amount of useful information available on the web that we can use to look at the principles of NMR (and then MRI). To start with, watch this video courtesy of (Professor Sir) Paul Callaghan, a Kiwi who literally wrote the book on microimaging (which is very high resolution MRI by another name):

 
It is a wonderful introduction to the basic concepts of nuclear magnetic resonance, the phenomenon we exploit for MRI. In fact, I would go so far as to say that this is the clearest explanation of NMR, by virtue of a stellar demonstration of the analogous effects of angular momentum (a spinning wheel) in the earth's gravitational field, that I've yet seen. Even if you've taken an NMR or an MRI course, it's worth watching. It's entertaining as well as informative.