Traditional radiology worklist systems often rely on rigid, rule-based logic that fails to account for radiologist fatigue, subspecialization, and individual workload. A study of 62 hospitals involving 2.2 million medical examinations found that these inefficient assignment patterns contribute to 17.7-minute diagnostic delays for expedited cases, resulting in financial impacts between $2.1M and $4.2M across hospital networks. To address this, AWS published a technical guide for deploying autonomous AI agents on Amazon Bedrock AgentCore that orchestrate clinical workflows by dynamically matching examinations to the most appropriate radiologist based on real-time data.

The proposed system utilizes a network of specialized AI agents to automate exam prioritization and radiologist allocation. The workflow begins when an exam is ingested into a picture archiving and communication system (PACS), triggering an intelligent orchestrator. This central agent coordinates with sub-agents like the Exam Metadata Synthesizer and Patient History Synthesizer to aggregate context. Simultaneously, the system employs imaging models, such as the Artery-aware network (AANet), to detect critical findings like pulmonary embolisms, which automatically elevates case priority. These agents leverage Amazon Bedrock foundation models to reason about assignment suitability, ensuring complex studies are distributed equitably to manage radiologist fatigue.

To maintain security, Amazon Bedrock Guardrails intercept queries to redact personally identifiable information (PII) before it reaches the orchestrator or is surfaced to users. The system incorporates two distinct memory strategies: short-term memory to maintain context during active sessions and long-term semantic memory to store assignment rationale and historical outcomes. Through continuous adaptive learning, the system identifies patterns from assignment rejections or SLA breaches, refining its decision-making logic over time. Integration with external tools—such as electronic health records and scheduling systems—is managed via the Model Context Protocol (MCP) through an AgentCore Gateway, enabling autonomous, secure data exchange for complex clinical orchestration.

Traditional radiology worklist systems often rely on rigid, rule-based logic that fails to account for radiologist fatigue, subspecialization, and individual workload. A study of 62 hospitals involving 2.2 million medical examinations found that these inefficient assignment patterns contribute to 17.7-minute diagnostic delays for expedited cases, resulting in financial impacts between $2.1M and $4.2M across hospital networks. To address this, AWS published a technical guide for deploying autonomous AI agents on Amazon Bedrock AgentCore that orchestrate clinical workflows by dynamically matching examinations to the most appropriate radiologist based on real-time data.

The proposed system utilizes a network of specialized AI agents to automate exam prioritization and radiologist allocation. The workflow begins when an exam is ingested into a picture archiving and communication system (PACS), triggering an intelligent orchestrator. This central agent coordinates with sub-agents like the Exam Metadata Synthesizer and Patient History Synthesizer to aggregate context. Simultaneously, the system employs imaging models, such as the Artery-aware network (AANet), to detect critical findings like pulmonary embolisms, which automatically elevates case priority. These agents leverage Amazon Bedrock foundation models to reason about assignment suitability, ensuring complex studies are distributed equitably to manage radiologist fatigue.

To maintain security, Amazon Bedrock Guardrails intercept queries to redact personally identifiable information (PII) before it reaches the orchestrator or is surfaced to users. The system incorporates two distinct memory strategies: short-term memory to maintain context during active sessions and long-term semantic memory to store assignment rationale and historical outcomes. Through continuous adaptive learning, the system identifies patterns from assignment rejections or SLA breaches, refining its decision-making logic over time. Integration with external tools—such as electronic health records and scheduling systems—is managed via the Model Context Protocol (MCP) through an AgentCore Gateway, enabling autonomous, secure data exchange for complex clinical orchestration.