Real-Time and Continuous AI Inference Workflow with PACS Integration

This study proposes an framework that automatically monitors incoming DICOM images from PACS,, applies multiple AI models, and stores the resulting annotations in standard DICOM formats (GSPS, SEG, RTSS). The system leverages an AI Worklist Manager to schedule and track AI tasks, ensuring seamless collaboration across various modalities without disrupting existing radiology workflows. The system’s successful deployment of (1) cerebral hemorrhage, (2) brain tumor segmentation,(3) hand joint scoring, and (4) breast analysis models for system validation demonstrated the system’s feasibility: AI results are directly available in the PACS viewer, reducing radiologists’ clinical burden and speeding up diagnosis. By adhering to IHE and DICOM standards, the framework ensures robust data exchange, full traceability, and minimal changes to established clinical processes.