Do you know if there’s a way of recreating one DTI direction volume from the others? We have a DWI data set with 200 directions. 3 of them are just pure noise, and another one has a few slices with ‘glitches’. For the moment, we thought about just dropping those directions using Mipav. But is there a way of recreating the directions from a weighted sum of the good directions? In the case of the 3 pure-noise directions it may not add much, but the direction with glitches still has a lot of good data in it…
Hello! One way to do this is to fit a model to the 197 directions you do have, and then use that model to predict the remaining three directions. We have shown that provided the model is a faithful model of the signal, at least in some cases, this would actually probably be more accurate than going back to the scanner to get another measurement of those three missing directions (http://journals.plos.org/plosone/article?id=10.1371/journal.pone.0123272). Many of the models in Dipy are equipped with a predict method that can do exactly this (predict out of sample directions).