Purpose or Learning Objective:
The main goal of this work proposes moving beyond isolated, task-specific AI tools in radiology toward a holistic Agentic AI orchestration framework. The contemporary radiology workflow is increasingly characterized by fragmented systems and high cognitive load, where radiologists must manually synthesize data across disconnected EMR, PACS, and reporting interfaces. These isolated models often lack the intuitiveness required for seamless clinical adoption, forcing radiologists to toggle between applications or validate findings without clinical context. Ironically adding friction to the diagnostic process rather than relieving it. The main goal of this work is to transcend these limitations by moving toward a holistic Agentic AI orchestration framework. By coordinating multiple semi-autonomous AI agents through a central orchestrator, each covering different parts of the diagnostic process, creates an interactive workflow with minimal human input. This model is designed to increase efficiency, reduce radiologist burnout, and improve patient care by automating repetitive tasks and closing gaps in communication and follow-up.
Methods or Background:
The framework breaks down the radiology workflow from image acquisition to follow-up. We propose the usage of these defined agents:
1) Triage Agent that analyzes incoming studies in real-time to prioritize critical findings (e.g., stroke, dissection, embolism) - It will attempt to prioritise the urgency.
2) Patient Summary Agent that interacts with the EMR and RIS to compile relevant clinical history and retrieve prior imaging results
3) Several Analysis & Measurement Agents that use existing, validated deep learning models for specific tasks like hemorrhage detection, tumor segmentation, and vessel characterization
4) Interactive Reporting Agent that creates preliminary structured reports by integrating findings from all other agents; and
5) Follow-up Agent that automatically parses, tracks, and manages recommendations for further imaging or clinical action.
A master Orchestrator Agent manages data flow. It integrates with PACS, RIS, and EMR. It also escalates uncertain results for human review. Comparisons were made with isolated AI tools in simulated high-volume settings.
Results or Findings:
Comparative simulations revealed that the Agentic AI framework significantly outperformed isolated AI toolsets and the radiology workflows. The holistic orchestration reduced overall diagnostic Turnaround Time (TAT), primarily by eliminating manual data aggregation steps.
Critical Prioritization: The Triage Agent will reduce "time-to-eyes-on-image" for emergent pathologies (e.g.,
intracranial hemorrhage, PE), ensuring high-priority cases bypassed standard queues.
Context Retrieval: The Patient Summary Agent will successfully retrieve and synthesize relevant priors and lab values, effectively removing the manual "chart-digging" burden.
Workflow Efficiency: By centralizing findings into a single preliminary report, the framework will reduce radiologist "context switching" the cognitive load of toggling between PACS, EMR, and separate AI viewers.
Follow-up Reliability: The Follow-up Agent will achieve capture of report recommendations, automatically logging them for tracking.
Qualitatively, radiologists report higher satisfaction scores due to the intuitiveness of the "pre-assembled" case context, which allows them to focus on decision-making rather than data hunting.
Conclusion:
Agentic AI in radiology advances beyond basic pattern recognition, creating a network of specialized, self-aware agents that improve workflow efficiency and radiologist satisfaction. While challenges like integration, regulation, and liability remain, this approach positions AI as a proactive, intelligent partner in patient care rather than just a tool.
Limitations: The most significant limitation is that the results are derived from "simulated high-volume settings" rather than a live clinical deployment. There may also be a regulatory and liability limitation of implementation of a Master Orchestrator in a clinical setting.
References:
1.Karunanayake N. Next-generation agentic AI for transforming healthcare. Informatics and Health. 2025;2(2):73-83. https://www.sciencedirect.com/science/article/pii/S2949953425000141
2.Banerjie S, Zhu Y, Freeman I, et al. Agentic AI in Healthcare: A Comprehensive Survey of Foundations, Taxonomy, and Applications. TechRxiv. Preprint posted online November 8, 2025. doi:10.36227/techrxiv.176238073.31262603
3.Tallam, K. (2025). From Autonomous Agents to Integrated Systems, A New Paradigm: Orchestrated Distributed Intelligence. arXiv. arXiv:2503.13754 [eess.SY]
4.Medical AI Agents: A Comprehensive Survey of Architectures, Cognitive Modules, and Clinical Workflows. TechRxiv. 2025. DOI:10.36227/techrxiv.176463029.99260745.
5.Li S, Xu J, Bao T, et al. A co-evolving agentic AI system for medical imaging analysis. arXiv [preprint]. 2025. arXiv:2509.20279
6.Xu G, Li X, Chen Y, Duan Y, Wu S, Yu A, Chiu C-H, Ni J, Tang N, Li T J-J, Yuille A, Jin W, & Shi Y. A Comprehensive Survey of AI Agents in Healthcare. TechRxiv [preprint]. 2025 Nov 14. DOI:10.36227/techrxiv.176240542.22279040
