SlideChat is a multimodal generative artificial intelligence assistant for whole-slide computational pathology across cancer types.

SlideChat achieves a 19.1% higher accuracy in interpreting gigapixel-scale whole-slide images across 31 cancer types, outperforming existing models in computational pathology.

• Why it matters: Accurate analysis of whole-slide images is crucial for clinical diagnosis and decision-making, yet current AI methods are limited to small patches and lack comprehensive interpretability.
• What they did: The team developed SlideChat by integrating patch- and slide-level encoders with a pretrained large language model, training on 274,233 multimodal instruction samples to connect WSIs with diagnostic reports.
• The result: SlideChat demonstrated superior performance on diverse question types and report generation, enabling improved diagnostic workflows, medical education, and clinical decision support.