Teaching fMRI analysis is a formidable task; one has to know some basic facts about statistics, image analysis, physiology, psychology, WW2 history, movie trivia, and a few pick-up lines to even get started. (The well-known line that uses "regions of interest" as a euphemism for erogenous zones has been quoted threadbare, but is still surprisingly effective.)
In spite of these challenges, image viewers can make it easier to learn these concepts. What I look for is an image viewer that is dynamic: one that can immediately update a time series by clicking on a voxel. Both AFNI and FSL have viewers that can show the time series in "movie mode," in which volumes at a voxel coordinate are displayed consecutively at a steady rate, like a flip-book.* This is useful for detecting things such as movement artifacts, and gives the student a better idea of how volumes are linked together to create datasets. When this is put together with the knowledge of TRs and voxel resolution, one begins to form a mental association between the sound of the magnet's gradient coils and taking images of the brain.
However, learning how a general linear model (GLM) is associated with fMRI data is hard for any newcomer. Although other concepts - such as k-space - can be equally difficult to learn, understanding the GLM is the heart of fMRI data analysis, and must be understood thoroughly before making an experiment. Most novices see the equation Y = Xb + e, have some idea that matrices are involved, and stop there. And why should they learn anything more? The rest of their training consists of knowing how to run scripts and look at the results, which requires little more than muscle memory.
FSLeyes, a new image viewer packaged with the latest version of FSL, allows one to see more clearly how the GLM is reflected in the data. In addition to the usual viewing functions of overlaying statistical maps on top of anatomical images, FSLeyes includes a viewing option called "FEAT mode," FEAT referring to FSL's modeling program. This opens up two additional windows: One listing the cluster coordinates for each contrast, and one showing how the model is fit to the data. You can also overlay the HRF for each parameter estimate individually, or overlay all of them simultaneously. By comparing this to the full model fit, it becomes clearer how the individual HRFs for each condition are added together, scaled, and then fit to the data.**
I recommend downloading the latest version of FSL (5.0.10 as of this writing) and experimenting with FSLeyes on sample datasets. Even if you do not use FSL to analyze your data, FSLeyes can be useful for building your intuition about how GLMs correspond to fMRI data, and for developing a better appreciation for how intricate and sometimes messy these models can be.
*I haven't found a comparable function in SPM; even though it can display time series data, you cannot see both the fMRI data and the time series change simultaneously. If someone know how to do this, please say so in the comments section below.
**AFNI can also do this, but it involves several more steps.