Hi Cristina,
This is interesting, because when there are no significant results, often the researcher becomes disappointed. But when there are too many significant results, such as throughout the whole brain, the researcher also becomes disappointed. So it feels as if one had to calibrate the results so that the researcher becomes satisfied in having some circumscribed regions about which stories could be said.
Far from trying to solve this problem here, if the design is correct and the images were all processed correctly, then it seems there is reasonable evidence that the behavioural variable is associated with FA throughout the whole brain. If this isn't something interesting, then perhaps consider using a global measure of FA (such as the global average) as an additional EV in the model.
Another possibility could be whether there could be another factor, other than age, sex, and education as already included in the model, that could be influencing both things, i.e,, both SDMT and FA, in your population. You'd have to try to identify this factor and include it in the model.
Hope this helps.
All the best,
Anderson