[reubens]

There is none. But if the multimodal model is exposed to enough medical knowledge, it may be able to interpret images without specific training

[rjtavares]

Labelling data is easier, I think. It will just take a while...