Paper tests evidence-linked multi-agent AI on biopsy reports

A new arXiv paper tests an evidence-linked multi-agent system for H. pylori case finding from gastric biopsy reports. In a 54-report de-identified pilot, the system produced 213 correct decisions out of 216 and linked answers back to source sentences. The sample is small, but the paper is worth including because it focuses on traceability, a key requirement for clinical AI workflows.

Key details: The paper was submitted to arXiv on July 7, 2026, It studies H. pylori case finding from gastric biopsy reports, The pilot used 54 de-identified reports, The system produced 213 correct decisions out of 216 and linked outputs to source sentences.

Why it matters: Small clinical-AI pilots are only useful if clinicians can trace the evidence, and this paper tests that evidence-linking requirement directly.