A missing feature specific of spinal cord compression ?

[etdufay]

Hello,

I wonder if I may have forget any distinctive feature of spinal cord compression.

Let me know if you have any information about such a feature.

I already have in my models the following ones: area, diameter AP, diameter RL, solidity, orientation, compression ratio, slice number and torsion.

I believe I could improve my prediction with one more feature.

Thank you very much !

Etienne

[jcohenadad]

LGTM! @valosekj ?

[valosekj]

Good point, @etdufay!

Three notes from my side:

1. torsion is calculated from orientation:

For that reason, we included only torsion (i.e., not orientation) in our original model.
However, torsion requires several slices for its calculation, which is a slight limitation for slices at the top and the bottom of the FOV --> it might be interesting to see a difference in model performance when using torsion vs orientation.

2. slice number is related to sci-zurich dataset, isn't it? Moreover, it can vary across subjects based on the number of slices. Thus, I wouldn't use it. Instead, vertebral level (e.g., C2) or intervertebral disc (e.g., C2/C3) might be more relevant. You can get this information from the vertebral labeling file.

3. In our original model, we included also age and height. Which were not found to be significant variables, though.

[etdufay]

1. Okay, thank you very much for these precisions about orientation and torsion !
   I'm going to benchmark the use of torsion vs orientation.

2. You're right, it deals with a zurich dataset, but with dcm-zurich (commit 3b2a27f8ab97ca5ca3409a4b7b1dbdc20e130dd3). I found out that it was relevant for the regression to set as an input variable the location in the spinal cord (compressions are more often in the first half of the slices). But indeed, what i did not realize, is that as MRI scans don't have the same number of slices, it is not interpreted properly by the regressor. Thus, I think I'm gonna replace the slice number by a number between 0 and 15 that points out a normed location in the spinal cord.

3. I think too that it will not impact upon the output.