0

I have a dataset of nifti files about Brain MRI performed to different patients and i've loaded them in Python using a library called Nibabel.

Here a slice of an MRI taken from my dataset:

enter image description here

As you can see this head is not perfectly aligned (in this case is slightly inclined to the right). Since i want to use this nifti dataset with a neural network, i would like to align properly every single Brain MRI.

I've found the tool Flirt (found in FSL) (I'm actually using a wrapper for Python), do you know if it can be used to solve my problem?

If yes, which are the appropriate settings for the tool ? It's the first time i work with nifti files :(

enter image description here

If this tool cannot be used to align Brain MRI, do you know tools/libraries i can use for this purpose ?

AleWolf
  • 133
  • 1
  • 6
  • 1
    Shouldn't your neural network work with non-perfect inputs too? – AKX May 20 '20 at 07:51
  • Thanks so much for your reply. It's the first time I use neural networks, in your opinion are they robust enough to tolerate this kind of misalignments? – AleWolf May 20 '20 at 07:53
  • 1
    Depends on what you're doing with the NN, of course. You may want to consider making any preprocessing (incl. this alignment) a part of the input stage of your NN program, so all inputs (training, test, validation, and finally inference) get the same treatment. (If they don't, your network will have bogus inputs for what it's trained for, and won't work.) – AKX May 20 '20 at 08:10
  • Yes, that is my idea – AleWolf May 20 '20 at 08:21
  • 2
    @AleWolf The topic you are asking about is called [image registeration](https://en.wikipedia.org/wiki/Image_registration), it's very useful as preprocessing before deep learning, it's not trivial topic, but you can use SimpleITK library in Python to do that, please see this [website](https://simpleitk.readthedocs.io/en/master/link_examples.html) which explains different methods to do `Image Registeration` – Bilal May 20 '20 at 08:37

0 Answers0