A new review outlines how intelligent agents could support sequential diagnostic workflows across slide overview, reasoning, and reporting stages.
Researchers from the Department of Pathology at Peking Union Medical College Hospital have published a review examining how intelligent agents could support pathology by modeling diagnostic processes as sequential and evidence-driven.
The review, published in the Medical Journal of Peking Union Medical College Hospital, examines how agents powered by large language models, vision-language models, and specialized computational tools could be organized around three connected stages of clinical pathology: low-magnification slide overview, diagnostic reasoning, and report generation.
While computational pathology has advanced from narrow machine-learning tools to models trained across various organs, diseases, and tasks, most current systems process images in a single pass. The researchers say this design overlooks how pathologists actually work, including searching across large tissue areas, changing magnification, and comparing multiple slides.
Three Stages of Agent-Based Pathology
The review highlights a roadmap for integrating AI agents into the clinical workflow:
- Overview Stage: Digital specimens are commonly stored as whole-slide images. Conventional systems divide these images into patches to highlight suspicious regions, but navigation agents instead treat slide review as a sequence of decisions, choosing where to move and when to zoom.
- Diagnostic Stage: Visual models can extract morphological features while large language models coordinate higher-level reasoning. These systems use chain-of-thought strategies to organize observed features and evidence. Retrieval-augmented generation can also bring guidelines, textbooks, and annotated cases into the reasoning process.
- Reporting Stage: Emerging systems attempt to combine information from several slides rather than generating text from a single image. Memory mechanisms may preserve previously observed features or retrieve similar historical cases, while dynamic adaptation can incorporate feedback from the pathologist.
Barriers to Clinical Implementation
The review identifies several major barriers to the adoption of these systems, including incorrect memories, model drift, privacy risks, and inconsistent outputs. The authors emphasize that the field has not yet demonstrated dependable performance across real diagnostic workflows.
Clinical usefulness will depend on transparent reasoning, traceable evidence, controlled updating, and thorough evaluation alongside practicing pathologists, the authors say.
Agent-based pathology could eventually reduce the time spent scanning large slides, help organize evidence from multiple specimens and tests, and support differential diagnosis. These systems may be especially valuable in complex oncology cases, rare diseases, and settings where specialist expertise is limited.
However, the authors frame this research as a roadmap rather than a clinically validated product. Future work must integrate modules across stages and test stability, reproducibility, data security, and clinical benefit in systematic trials before pathology agents can become components of routine diagnosis.
ID 25712658 | Lab © Xunbin Pan | Dreamstime.com
