Choosing the Statistical Test (if exists)

#1
Hi! I'm new here and I need help with this:

The data:
I have 1000+ subjects each one scanned with a MR 3 times (with different sequence T1, T2 and IR). Each scan yields (after a segmentation) to the 15 volumes of the brain region in study (the brain region has 15 "sections" which volumes' are interesting). I also have age, sex, and other variables of each subject.

The goal:
One of the goals of the project is to do a comparison between the segmentation results coming from 3 sequences (but I have no ground truth, I only have kind of guidelines of how the volumes behave).

From the evaluation criteria (the guidelines about volumes' behavior). I can distinguish 3 types:
  • Comparison between two groups: Group A has smaller volumes than group B.
  • Comparison of a group throughout time: Volumes decrease with age
  • The two previous at once: Volumes from group A decrease faster with age than volumes from group B.
I also have other criteria not related to subjects. For example, section #3 (one of the 15) is the biggest one.

How would you know which sequence yielded to better results?
Is there a statistical analysis method that compares things like this?

Thank you!
 
Last edited:

hlsmith

Not a robit
#2
Can you provide a cross-sectional slice of your data so we can better understand what you are working with. You can fiddle with the data a little to deidentify it if needed.
 
#3
Thank you for the fast reply.
First of all I must say I am an engineer and I never been taught this (don't know why actually). I don't even know what is cross-sectional data. From internet I see that is observations made in the same point in time (?)

I have to clarify. The 3 scans of the subjects are only made once. When I say years is because some subjects are older than others, but all just had 3 scans at the same time.

That said, I don't know if it'll be helpful, but I have 4 CSV files with volumes:
  • Volumes coming from T1 sequence
  • Volumes coming from T2 sequence
  • Volumes coming from IR sequence
  • Data's subjects
Don't know if this is what you asked for... I add a screenshot of how are they formatted.

1559929249345.png