Probably. I mis-spoke.

Generally speaking, yes, but standard LLMs now have their own built-in image recognition so we might not be too far off from LLMs being able to describe to the doctor what they should look at on an image that's been flagged. Take a capable LLM of a few hundred billion parameters, train it on tons of medical images and such, and it might actually be helpful. Or it might not, no way to find out until tested.