WSI-LLaVA: A Multimodal Large Language Model for Whole Slide Image