How to deal with animals with miniscope and cable

Hi, I’m using DeepLabCut to analyze animal behaviors in open field arena when the calcium imaging was recorded simultaneously. Currently I’m able to successfully label the animal position. The issue is that I want to get mouse behaviors in details like left turn, right turn, rear, rest, and so on. But the cable connected to mouse and the miniscope disrupt the recognition of GUI (sometimes my naked eyes can’t find the snout or ear from the sample images). I have two questions:

  1. how could I deal with head mounted scope and cable, would you share some experience with me?
  2. As for the behavior details, which body parts should I label in order to train the GUI for getting rear and turns?
    Thank you in advance.

Hi @Lingjie, welcome to the forum! A very simple tip is to only label what your eyes can see. If the cable occludes a body part in some frames, do not label it (your model may otherwise learn features that correspond to the cable—which you do not want—rather than the actual body part you would like to reliably track). As for behavioral analysis, although 2 body parts defining a line would be enough to distinguish left/right turns, I would encourage you to label more (e.g., ears and along the back to characterize gaze direction). Hope that helps!

Thank you so much for the answer. A follow-up question on labeling: If snout or one ear is occluded by miniscope in some frames, can I deal with it in a similar way as the cable? Just skip it and the model will learn it?

Absolutely, do not label guesses :slight_smile:

Thanks a lot :smile: