Yes, but it's not a trivial thing to do. I'd put my money with 'ordinary' computer vision algorithms if you want that sort of stuff.
All 3d scanning algorithms have several tradeoffs; Complexity, robustness, number of frames required, processing time, accuracy and resolution.
In order to get scans that are good enough to detect under extrusion, you would need a pretty awesome resolution and accuracy. It would also need to be extremely robust, as false positives / negatives are a big no-no. This usually means that you require either a ton of frames and/or a ton of processing time and or huge complexity. But then again, even a ton of frames is kind of a no-no, and the same can be said for a ton of processing time, leaving only one thing to be unoptimized; The complexity.