Accuracy for ROI extraction

[tomDag25]

This issue is opened to discuss the way we will compute the accuracy of our ROI extraction.
We will compute the presence or the absence of the label of the radiologist in our ROI extracted volume.

A positive will be when the point is in the volume and a negative will be when the point is not.

[tomDag25]

The difficulty to compute this is that the points are given in the DICOM slices and therefore have only two coordinates in the sagittal plan whereas our ROI are extracted directly in 3D.

[abelsalm]

Also, I just note it here but the real interest of this pipeline coudl also be for us to fine tune the values we're using to extract the patches, having a way to minimize the number of slics and optimize the accuracy. For instance may be for NFN especially this could be improved!
Once you're done @tomDag25 I can run those tests on each pathologies and try to optimize a bit that!