I need a quality control tool to filter quality MRI from those having artifacts (motion artifacts / ghosting blurring, noise, atypical anatomy, etc.).
If I remember last year’s fast.ai course, just training a CNN to classify MRI by artifacts will not generalize well to handle unknown artifacts.
And there is no way for CNN to return a “distance from normal MRI” which could be threshold to filter good MRI from bad ones.
Am I correct? Or is there a way to make such binary classifier (predicting good vs bad MRI) ?
I do not have a good training database, but anyway I am interested to know what’s possible in theory.)
From what I read in the literature, there is no simple way to do QC with CNN, but I have seen interesting projects like CNNArt
Examples of MRI artifacts from the UK Biobank study