As I described in previous posts, the first version of iBEAT will take you from a raw T1/T2/FA image to an image that has been skull-stripped and labeled according to the AAL atlas. You can then use the labeled brain to extract data from regions of interest (ROIs) in another software, such as FSL. Essentially, iBEAT is a user-friendly way to register an infant brain to a template.

The problem with user-friendly software is that when something goes wrong, it can be difficult to determine why or how.

As I was looking through the volume measurements I got for each of the AAL ROIs using FSL, I realized that for about half of our subjects, the volume measures were very similar, but they were not identical. As I was extracting the data from FSL, I checked every file to make sure that I was not mistakenly copying the same subject data into multiple folders. If the numbers were one or two digits off, I recognized that it was not an identical file, and moved on.

Unfortunately, it wasn’t until all the data was in a spreadsheet together that I realized that there were subjects where every volume measures was only one or two digits off from other subjects. For example, S1001 had volume measures of 13001, 14534, 13256, etc. and S1002 had volume measures of 13000, 14535, 13255, and S1003 had volume measures of 13002, 13534, 13254. This was true for about half of the subjects. The other half had plenty of variability in their data, which made it tricky to identify these patterns.

Once I noticed the pattern, I went back to check that I had copied over the right files and extracted the right data from FSL. Everything seemed fine. Then I opened the labeled brain files from two subjects with very similar data. The images looked nearly identical. Then I opened their tissue-segmented brains, and they were clearly different brains. Something had gone wrong between the tissue-segmentation step and the brain labeling step.

Because iBEAT uses a GUI, and does not have an option for scripting, it was difficult to determine how to solve the problem. As a starting point, I decided to re-run a few of the tissue-segmented brains through the brain-labeling step to see if it might work correctly this time. It did. For three of the seven brains I needed to correct. The other four still looked nearly identical.

I processed the remaining four brains through iBEAT again from the beginning to see whether we could fix the results that way. It worked. For two of the four brains. For the remaining two, I was losing hope, so I just tried clicking the “label brain” button again. I still got the error. Then I tried one more time. This time, the brains were labeled correctly and matched the tissue-segmented brains.

Clicking “label brain” gave me incorrect results 3 times, but correct results the 4th time. I have no explanation for this, and when looking at the FAQs for iBEAT, I noticed that the solution to other errors in iBEAT is “try running it again.”

I hope that anyone reading this can learn from my mistake. Always check and double check your results. And if you are using iBEAT and something goes wrong, try running it again. The results may change.

*As a footnote, I would like to add that it concerns me that my results changed when I did the same thing multiple times. The problem with the GUI is that I have no idea what is going on behind the scenes, and I just assume that clicking the same button will always produce the same results. I don’t know why it does not for iBEAT, and it makes me uneasy to use a software that is not transparent; however, this is the only software we have found that makes the preprocessing of infant MRI data feasible without a large number of manual edits or errors.