An AI image classifier tuned on labeled data seems like a possibility and a good area of research. It would take a large dataset of imagery and computation to train an open computer vision model.
GPTs, though, are AI agents that only operate within ChatGPT plus for subscribers. It is not an externally accessible product like API AI models from OpenAI. GPT-4-Vision also is not trained for this type of medical diagnosis.
It would be possible to make a GPT that has actions that interface with an API where the external trained vision model is located, but there wouldn’t really be much value added by the AI’s interpretation of vision results.