CyberNeuro launches WandaMind: AI agents for private neuroimaging analysis
Tackles 40% to 69% accuracy boost in NeuroBench while slashing token costs by 89%.
A 13-author team from the University of North Carolina and collaborators introduced CyberNeuro, an open-source agentic workbench designed to automate end-to-end neuroimage and clinical data analysis while preserving clinical-grade privacy. The system pairs a custom local LLM, WandaMind, with four specialized agents orchestrated through a secure MCP bridge and pinned execution layer. Researchers can now run complex cohort-scale pipelines using plain-language prompts, eliminating months of manual curation, pipeline execution, and post-processing QC.
In benchmarks on the public NeuroBench suite, CyberNeuro boosted held-out domain accuracy from 40% to 69% over the baseline and completed a 10-batch cohort workflow using only 10.6% of tokens via WandaMind compared to 100% reliance on cloud providers in prior systems. A human-in-the-loop verification panel ensures biomedical-grade quality control, bridging AI efficiency with clinical rigor.
- CyberNeuro uses WandaMind LLM + 4 agents (Planner, Validator, Dispatcher, Reporter) for secure, natural-language neuroimaging workflows
- Accuracy on NeuroBench jumps from 40% to 69% and token usage drops to 10.6% locally vs. 100% cloud in prior systems
- Production-ready modules available via open-source release; human-in-the-loop QC panel enforces clinical standards
Why It Matters
Democratizes large-scale neuroanalytic pipelines for resource-limited labs while keeping data private and compliant.