Anatomical Mapping Question

[ajulian3]

When performing a quantitative analysis on a brain, how does anatomical mapping information change in terms of interpretation when there is a physical deformity within the subject. Specifically, how does one go about standardizing the neural network data when a physical deformity is present?

[adjordan]

I believe I can comment on this knowledgeably, although I hope someone else can give a more complete answer. It is very difficult to standardize data from a damaged area of the brain because it is not always clear what effect the damage is having on the data. The damaged area will have an effect on it's surrounding connections, geographically and functionally, so it is taken into account slightly, but it's impossible to tell how the damage is actually affecting the data to the extent that would allow you to correct it. This is my understanding, so I hope someone else adds more to this!

[akim1]

The brain, anatomically, isn't really that different from individual to individual. I guess it really depends on the physical deformities, but all brains have characteristic landmarks (e.g., presence of different sulcus). You can always use these landmarks as reference points to determine where on the brain you actually are. If a piece of a brain is completely absent, then, it would just need to be compensated for with other landmarks.

[yaxigeigei]

Experimentally, if the question of interest does not involve the damage, we can simply avoid such data. However, if it does or we are dealing with clinical applications. One possibility I can imagine is that we ignore regions which have too much variance given a threshold and comparing the difference of other regions or even global properties. I would say it really depends on specific case and problem.

[whock]

In general, there is a certain amount of anatomical variation in brain structure from one individual to the next due to abnormalities but also just individual differences. so although a region such as visual cortex V1 has a set of (x,y,z) coordinates saying where it is, the exact location in space is different enough from subject-to-subject to be a thing.

A common solution in neuroimaging is to use a standard template of an "average" brain and fit each individual's brain to it in a process called image registration and normalization (this also corrects for the subject fidgeting inside the scanner).

Here's a PDF of fmri preprocessing steps as an example of what kinds of variations there are from one scan to the next (not all due to anatomical differences) and what can/can't be done about them:

https://ngp.usc.edu/files/2013/06/Lei_Preprocessing_08_11_2011.pdf

Will