Framework separates data analysis from trustworthy scientific reporting
CONFIRM: Governing Scientific Reporting for Trustworthy Agentic Neuroimaging
Artificial IntelligenceMultiagent Systems
Summary
Scientific studies in brain imaging sometimes report findings that aren't reliable because of hidden problems like confusing variables or weak data. The authors showed that current AI tools can mistakenly accept these flawed findings. They created CONFIRM, a system that carefully checks hypotheses before analyzing data and uses strict rules to decide which results to share. This helps avoid false claims and provides guidance when something isn't right. Their new benchmark, NeuroClaimBench, tests how well such systems work with real and tricky example studies.
What this means in practice
- •For data analysis teams: Improve automated reporting of brain imaging study results by filtering claims through fixed analysis plans and strict verification gates.
- •For healthcare ai developers: Enhance trustworthiness of AI tools that autonomously analyze medical brain scan data by reducing false positive findings.
Authors
Yuxiang Wei, Lijing Zhu, Qizhen Lan, Yanteng Zhang, Vince D. Calhoun
Abstract
Large language model (LLM) agents can autonomously run neuroimaging analyses from research questions to reported findings. Existing agentic systems use tools to check that code runs and reported values match its output, but rely on LLM judgment to decide which results warrant scientific reporting. We show that LLM judges endorse site-confounded findings even when the confound is presented as evidence. We present \textbf{CONFIRM}, a framework that separates running an analysis from deciding which findings to report. CONFIRM encodes each hypothesis as a machine-readable, pre-specified analysis plan that fixes the claim and its pass criteria before computing results. It then filters results through five deterministic gates (search history and multiplicity, confounds, power, analytic stability, and replication) under three configurable reporting standards, from least to most demanding: \textit{Screen}, \textit{Replicate}, and \textit{Full}. Failed claims receive diagnoses that guide follow-up analyses. To evaluate reporting decisions at the claim level, we introduce \textbf{NeuroClaimBench}, comprising 489 executable neuroimaging claims with frozen data, including literature-supported findings and 150 controls designed to violate individual methodological requirements. On NeuroClaimBench, CONFIRM falsely supports at most 3\% of controls, compared with 23--38\% for the baselines, while recovering up to 61\% of literature-supported claims. When auditing an existing agent, CONFIRM reduces false support on controls from 23\% to 0\% while retaining the agent's literature-supported claims. Its diagnostic feedback also produces more supported revisions than Self-Refine with fewer LLM calls. Together, CONFIRM and NeuroClaimBench show that basing reporting decisions on pre-specified analysis plans and deterministic gates is essential for trustworthy autonomous scientific discovery.