Evaluating the Diagnostic Performance of Large Language Models on Complex Multimodal Medical Cases | Synapse