How to make 3d approach for different depths?

I am working on a dataset that has spinal cord ct scans taken in the axial plane. The patient folder has slices of the CT Scans. It is a multi-label classification problem. I need to find the types of fractures that happened. Correct me if I’m wrong, but I think here the 2d approach doesn’t work well because each slice will not have whole spinal cord since it is taken in the axial plane. So, I wanted to try 3d approach by stacking the slices together. But the problem is that each patient is not having same number of slices. On average they are having around 300 scans. The highest is in 1000 and the lowest is in the 50’s. How should I proceed now?

Hi, @Raviteja !

It depends on the distribution of those number of slices. As a starting point I can think about a couple of possible solutions:

  1. Pad with black images the patients that have less than 1000 scans until reaching that number.
  2. Decrease the dimension of the examples that have many scans with some kind of pooling.
  3. What I think would be better, a combination of both.