Yes, it's possible to use transfer learning leveraging trained DL CNN models in the public domain (e.g. Keras) using as inputs the 3D images with 10 Channels (rather than 3 -- R-G-B).

I recommend checking out this tutorial which does a good job to explain how to use transfer learning with inputs of different sizes and channels (i.e., 1 channel and n channels) --> https://www.youtube.com/watch?v=5kbpoIQUB4Q