A Multi-modal Agentic Co-pilot for Evidence Grounded Computational Pathology

PathPocket is a multimodal AI co-pilot system designed to assist pathologists by grounding diagnostic recommendations in verifiable medical evidence. Built on a comprehensive pathology knowledge base of 110,472 documents and 4.55 million entities, the system demonstrates significant improvements in diagnostic accuracy and pathologist confidence across 200,000+ real-world cases.

PathPocket represents a meaningful advancement in clinical AI by addressing a critical gap between AI capabilities and evidence-based medicine practices in pathology. Traditional clinical decision support systems have struggled to provide transparent, verifiable reasoning for their recommendations. This system solves that problem through a multimodal architecture that processes text, microscopy images, and whole-slide imaging while maintaining explicit links to supporting medical literature.

The construction of a 110,472-document pathology evidence corpus organized by evidence hierarchy demonstrates rigorous methodology—distinguishing between clinical guidelines, research findings, and expert opinions. This foundational work provides the infrastructure necessary for trustworthy AI integration into medical workflows where liability and accuracy directly impact patient outcomes. The resulting hypergraph of 4.55 million entities enables sophisticated multi-agent reasoning that retrieves, filters, and synthesizes evidence before generating diagnoses.

The practical impact extends beyond incremental performance gains. User studies showing improved diagnostic confidence among pathologists suggest the system functions as a genuine decision-support tool rather than a black-box recommender. This collaborative approach—where AI augments rather than replaces expert judgment—addresses adoption barriers that have limited previous clinical AI deployments. The system's ability to handle gigapixel whole-slide imaging represents technical sophistication that mirrors real-world diagnostic complexity.

Future applications likely include integration with hospital information systems, expansion to other medical specialties, and regulatory pathway clarification. The open publication of methodology through arXiv suggests potential for academic adoption and refinement, though commercialization timelines and reimbursement models remain undefined.

• →PathPocket grounds AI pathology recommendations in verified medical evidence from 110,472 structured documents, addressing transparency requirements for clinical adoption.
• →Multi-agent reasoning framework processes diverse inputs including text queries, region-of-interest images, and gigapixel whole-slide imaging within a unified system.
• →User studies demonstrate pathologists achieve improved diagnostic accuracy and confidence when using PathPocket, validating human-AI collaboration over autonomous diagnosis.
• →System performance significantly exceeds existing state-of-the-art across 200,000+ real-world clinical cases, establishing new benchmarks for computational pathology.
• →Evidence-grounded approach enables verifiable diagnosis chains that improve clinical liability profiles and support regulatory compliance in medical AI deployment.