Get started
Work

CyberBench: Agentic AI playbook for PRIMED-AI

Year: 2026
Type: other
Abstract: Translating clinical AI into routine patient care fundamentally depends on trustworthy, AI-ready imaging data. Yet, generating these datasets remains out of reach for clinicians, who must rely on specialized technical personnel and ad-hoc pipelines. This operational bottleneck persists because current image processing workflows depend on application-specific software environments and lack efficient, standardized quality control. Taken together, these barriers impede the scalable generation of reliable imaging traits, driving a profound gap between surging AI innovation and real-world clinical utility. Without standardized, auditable data curation protocols, clinical AI models will remain brittle, unscalable, and difficult to deploy across diverse healthcare systems. This project addresses these rate-limiting barriers by introducing a scalable, agentic AI-driven community playbook built upon our field-tested software infrastructure. Our prior work has demonstrated the transformative capacity of agentic AI to bridge the divide between raw clinical archives and deployment-ready applications by converting expert human workflows in image processing, quality assurance, and execution management into autonomous, LLM-invocable skills. Validated across 17 multi-site cohorts comprising over 123,000 subjects, our agentic AI framework can reduce image processing timelines from 2–3 months to less than a week. Building upon the proven performance of our agentic AI architecture and enthusiastic user adoption, we will develop and deploy portable standard operating procedures (SOPs) paired with executable reference implementations. To realize this vision, our operational strategy delivers three interconnected pillars across the PRIMED-AI ecosystem. First, we will establish an integrated clinical imaging data curation and quality control framework that leverages hierarchical multi-agent orchestration to issue auditable data quality certificates for downstream deployment of clinical AI tools. To ensure long-term scalability, the framework will dynamically learn from existing knowledge of image processing documentation and auto-ingest emerging algorithms from literature via specialized skill-creation modules. Second, we will extend the agentic AI architecture into a multimodal scientific discovery engine that connects imaging features with genetic and health record trajectories under strict, pre-built quality guardrails. Third, we will deploy an interactive educational framework that pairs agentic tools with competency-based curricula to train clinical end-users in AI co-pilot operations. Developed through iterative feedback loops with clinicians and PRIMED-AI consortium partners, these portable frameworks will de-risk project feasibility and establish the vital infrastructure required for real-world clinical AI adoption. By converting static imaging archives into privacy-preserving, regulation-ready, and teachable digital assets, this research will accelerate the delivery of reliable, AI-assisted precision medicine to improve diagnostic precision and health outcomes across diverse patient populations.
DOI: 10.48321/d12962bf31
Author: Guorong Wu
Language: English
Field: Medicine
Open Access status: green
OpenAlex ID: W7214047602
Cites:
Cited by:
Related to: