NeuroAIHub: An AI-Driven Framework for Automated Curation and Discovery of Neuroradiology Datasets
Benyamin Gheiji1, Sina Moradi1, Mahsa Vatanparast1
1From the Mashhad University of Medical Sciences (B.G., M.V., D.E.), Mashhad, Iran; Zumud, Zumud inc. (S.M.), Twickenham, London, England, UK; Tehran University of Medical Sciences (S.S.), Tehran, Iran; Research Center for Noncommunicable Diseases (M.A.B.), Department of Immunology (M.A.B.), Jahrom University of Medical Sciences, Jahrom, Iran; Department of Medicine (O.I.A.), Shenyang Medical College, Liaoning, Shenyang, China; Lorestan University of Medical Sciences (S.G.), Khorramabad, Iran; Shahid Beheshti University of Medical Sciences (M.H.-F.), Tehran, Iran; Department of Radiology (J.D.R.), University of California San Diego, San Diego CA, USA; Department of Radiology (J.D.R.), Scripps Clinic Medical Group, San DIego CA, USA; Department of Radiology (E.C.), Duke University Medical Center, Durham, NC, USA; Department of Radiology and Neurosurgery (R.J.), NYU Grossman School of Medicine, New York, NY, USA; Radiology Informatics Lab (M. M.), Department of Radiology, Mayo Clinic, Rochester, MN, USA and Department of Radiology (S.F.), University of Pennsylvania, Philadelphia, PA, USA.
Abstract:
Neuroradiology datasets hold significant potential for advancing neuroimaging research, yet identifying relevant and up-to-date resources remains challenging. NeuroAIHub is an artificial intelligence (AI)-driven framework designed to automate dataset discovery, improve accessibility, and support structured exploration of neuroradiology datasets. A foundational database was constructed through a multi-reviewer extraction process with standardized metadata harmonization. To maintain and expand the registry, an AI-based updating workflow performs monthly web searches, extracts structured metadata from heterogeneous sources using large language models (LLMs), and submits candidate entries for developer validation before integration. An LLM-powered conversational agent enables natural-language dataset retrieval, analytical queries, and visualizations, while a structured web interface supports reproducible filtering. NeuroAIHub currently hosts 180 datasets across six diagnostic domains and is available as a web application (https://neuroai.streamlit.app/) and open-source Python package (https://github.com/NeuroAIHub-Registry/NeuroAIHub; https://pypi.org/project/neuroaihub). The platform provides a continuously curated resource designed to improve reproducibility, transparency, and efficiency in neuroradiology dataset discovery.
