PathoSage: Towards Multi-Source Evidence Adjudication in Pathology via Experience-Aware Agentic Workflow
PathoSage is a new AI framework that improves pathology analysis by separating evidence collection from decision-making, reducing hallucinations in multimodal large language models. The system uses structured evidence deliberation and a reliability-tracking mechanism to better judge conflicting medical information, outperforming existing pathology AI models.
PathoSage addresses a critical limitation in applying large language models to medical pathology: the tendency of MLLMs to fabricate morphological features and struggle when handling conflicting diagnostic evidence. The framework's three-stage architecture—separating knowledge retrieval, evidence collection, and adjudication—represents a meaningful shift in how AI agents should approach high-stakes decision-making domains. This separation prevents context contamination where earlier conclusions bias later analysis, a common source of error in chain-of-thought reasoning systems.
The research emerges against a backdrop of increasing AI adoption in healthcare, where the stakes of model errors are particularly high. Previous pathology systems merged multiple evidence sources into shared contexts, creating vulnerability to conflicting information. PathoSage's innovation lies in its Beta-Bernoulli experience system, which continuously tracks tool reliability without requiring retraining. This allows the framework to learn which diagnostic tools are trustworthy over time and weight future decisions accordingly.
For the broader AI and healthcare sectors, this work demonstrates that agentic systems require more sophisticated evidence processing than simple tool integration. The explicit conflict analysis and anchoring-bias mitigation suggest that multi-agent reasoning systems benefit from adversarial or independent evaluation stages rather than naive merging of outputs. Institutions developing diagnostic AI systems may adopt similar deliberation architectures to improve reliability and transparency.
Looking forward, the pathology AI community will likely focus on validating PathoSage in clinical settings and extending similar evidence adjudication frameworks to other medical specialties. The reliability-modeling approach could become standard in regulated healthcare AI where explainability and robustness are prerequisites for deployment.
- →PathoSage separates evidence collection from adjudication to prevent biased decision-making in pathology AI
- →The framework includes a reliability-tracking system that continuously assesses tool performance without retraining
- →Structured evidence deliberation reduces hallucinations and handles conflicting diagnostic information more effectively
- →The approach outperforms existing multimodal large language models and agentic baselines in pathology reasoning
- →Explicit conflict analysis and fresh-context evaluation mitigate anchoring bias in high-stakes medical decisions