VERITAS: Verifiable Epistemic Reasoning for Image-Derived Hypothesis Testing via Agentic Systems

A research team from the Technical University of Munich has introduced VERITAS (Verifiable Epistemic Reasoning for Image-Derived Hypothesis Testing via Agentic Systems), a breakthrough multi-agent AI framework designed to automate clinical research workflows. The system addresses the fragmented process of drawing conclusions from multimodal medical data—which typically requires coordinating expertise across clinical specialties, radiology, programming, and biostatistics—by decomposing the workflow into four phases handled by role-specialized AI agents. VERITAS autonomously tests natural-language hypotheses on clinical datasets while producing a fully auditable evidence trail, meaning every statistical conclusion can be traced through inspectable, executable outputs from analysis plans to segmentation masks to statistical code.

In evaluation, VERITAS demonstrated 81.4% verdict accuracy using frontier models and 71.2% with locally-hosted open-weight models (8-30B parameters), outperforming all five single-model baselines. The system was tested on a tiered benchmark of 64 hypotheses across cardiac and brain glioma MRI datasets (ACDC with 150 subjects and UCSF-PDGM with 501 subjects). Crucially, VERITAS produced the highest rate of independently verifiable statistical outputs at 86.6%, ensuring even failed analyses remain diagnosable through artifact inspection. The framework introduces an epistemic evidence label system that mechanically classifies outcomes as Supported, Refuted, Underpowered, or Invalid by jointly evaluating significance, effect direction, and study power—a critical distinction in medical imaging where non-significant results often reflect insufficient sample size rather than absent effects.