Brain AI Scientist: An Evidence-Based Multi-Agent System for Translational Neuroimaging in Precision Psychiatry

arXiv:2611.18634v1 Announce Type: cross
Abstract: Resting-state functional magnetic resonance imaging (rs-fMRI) has revealed widespread brain-network alterations in psychiatric disorders, yet translating these findings into individualized clinical decision support remains challenging. Here we introduce Brain AI Scientist, an autonomous multi-agent framework that integrates patient-specific fMRI, clinical information, and scientific evidence through retrieval, verification, and prioritization. Guided by Research Domain Criteria and evidence-based medicine, it builds traceable chains linking phenotypes, anatomical abnormalities, biomarkers, mechanisms, interventions, and validation strategies. In a representative autism case, Brain AI Scientist generated 37 anatomical hypotheses, retained 26 after literature verification, and derived 65 evidence-supported biomarkers from 90 unique publications to generate patient-specific treatment and validation guidance. Across 17 autism spectrum disorder cases, it outperformed GPT-3.5 baselines with and without fMRI impression in evidence grounding, personalization, clinical actionability, and biomarker-treatment reasoning. Brain AI Scientist therefore provides an auditable framework for translating patient-specific neuroimaging into individualized clinical decision support for precision psychiatry.
This article has been indexed from cs.AI updates on arXiv.org