The PathChat AI assistant, built specifically for human pathology, integrates a proprietary vision encoder trained on a corpus of vast histology images with a fine-tuned Large Language Model. Faisal Mahmood and team present this breakthrough assistant that diagnoses images provided with clinical context.

• PathChat’s vision encoder is adeptly pretrained on over 100 million images across diverse patient cases.
• This AI attains 87% diagnostic accuracy utilizing open-ended questions and expert evaluation.
• Outperforming other multimodal AIs, PathChat generates accurate responses that impress pathologists.
• As a versatile AI, it holds potential for advanced uses in pathology education, research, and clinical decision support.

PathChat epitomizes the success of combining deep learning with specialized domain knowledge, yielding an assistant that could significantly benefit the medical field by augmenting the workflow of human pathologists and contributing substantially to medical training and research.