I plan to use ADNI brain MRI dataset whose data are in Nifti. SPM 12 has been used for segmentation (grey matter-white matter). I can visualize them but could not understand how to use them in deep learning with the correct format. Should I convert them to jpeg or png (I read a few articles that use them as jpeg)?
The second and most important question is how can I determine which slice of the file should be used? Should it be the middle of them? Nifti files I used have 160 slices for each subject's brain and I did not see any article that mentions how to select slices.
Can you give some idea and recommend resources to understand the fundamental of the answers, please?
it depends on your application, you can use jpeg format if u want use cnn or you can extract timeseries in terms of using rnn.
for the second question i think it depends on brain regions that are related to your analysis. my point is you should have a prior knowledge about the activations and then select the corresponding slices for feeding your network.
I would suggest you check out the Toolkit NiftyNet. It directly takes nifti files as input (still you need to preprocess it) and extracts image volumes/slices from the same locations of different modalities, so no need to convert it to png or something.
As for the second question, I think it really depends on the problem you're working on, or the regions you're interested in. Let's say if you're working on memory tasks, then maybe hippocampus would be your region of interest, which would be the input of your model. NiftyNet provides several sampling schemes which should be enough.
you should directly use nifti files as an input.
And you try to get the samples from each slice of the 3d volume. Then use some method to get the balanced features from each class. (from background and foreground)
if you don't have any problem with the labels of images. you can use this source to convert .nii files to .png format.
this source is for python and Matlab.
https://github.com/alexlaurence/NIfTI-Image-Converter
- Badges
- Science topic