The abstract is read at the publisher; the summary is JClub's.
NeuroImgAgent, a cross-species fMRI multi-agent platform, corrected 111 of 120 induced orientation errors with GPT-5.5 and improved usability to 92.5% in 120 paired challenge cases.
- Why it matters: Cross-species fMRI requires coordinating complex, species-specific workflows and reliably judging intermediate outputs to avoid silent failures. Evidence-gated quality control is needed to decide when to proceed, recover, or stop.
- What they did: The platform uses large language models to translate natural-language requests into workflows for humans, macaques, rats, and mice, with quantitative checks and vision-language model assessments gating processing. It was evaluated on 268 paired cases versus fMRIPrep, AFNI, and RABIES, and on NeuroAgentBench and 120 paired challenge cases with blinded expert usability ratings.
- The result: Preprocessing metrics were broadly comparable to established pipelines across 268 paired cases. In induced-error tests, GPT-5.5 corrected 111/120 orientation errors and achieved high task success on NeuroAgentBench, while full configuration raised usability from 76.7% to 92.5% and reduced unusable outputs without warning from 18.3% to 4.2% versus quality-control ablation.
The findingWhy it mattersWhat they didThe result
AI summary of the abstract · may contain errors · not medical advice · read the paper.