AI Integration for Intelerad Ophthalmology PACS

AI integration for Intelerad Ophthalmology PACS connects at three key functional surfaces: the worklist manager, the image viewer, and the reporting module. The primary workflow begins when a new ophthalmic study—such as an OCT, fundus photo, or visual field test—is archived into the PACS. An automated DICOM listener or a scheduled task can trigger an AI inference service via a secure API call, sending the anonymized images for analysis. Results, including structured findings like retinal layer thickness maps, DR severity scores, or glaucoma progression flags, are returned as DICOM Structured Reports (SR) or JSON payloads and attached to the original study. This enriches the study metadata before the ophthalmologist even opens the case, allowing the worklist to be prioritized by clinical urgency.

Within the reading session, AI findings are presented as interactive overlays or a side panel within the Intelerad viewer. For an OCT scan, an AI-powered segmentation overlay can highlight fluid volumes or retinal layers directly on the B-scan. For a fundus image, bounding boxes and confidence scores for pathologies like microaneurysms or hemorrhages are displayed. This integrated presentation allows the clinician to rapidly verify AI suggestions, correlate findings across modalities, and incorporate quantitative data directly into their narrative report using context-aware macros or auto-populated structured report templates. The integration is designed to be non-disruptive, preserving the native Intelerad user experience while augmenting it with AI-derived intelligence.

A production rollout requires careful governance. A phased deployment typically starts with a silent mode, where AI runs in the background and results are logged but not displayed, to establish baseline performance and clinician trust. Following validation, a concurrent read mode is enabled, where AI findings are visible as a secondary opinion. Key technical considerations include managing GPU-accelerated inference for high-volume OCT analysis, implementing RBAC to control which user roles see AI annotations, and ensuring all AI activity is logged to the audit trail for compliance and MIPS reporting. The final architecture should treat the AI service as a stateless, containerized microservice, communicating with Intelerad via its published REST APIs and DICOMweb endpoints, ensuring scalability and ease of maintenance across a health system's imaging network.

The integration architecture connects to Intelerad's PowerReader Ophthalmology workstation and its underlying Vendor Neutral Archive (VNA). The primary data flow is triggered when a new ophthalmic study—such as an OCT, fundus photo, or visual field test—is stored. Using a DICOMweb listener or a workflow manager API hook, the system pushes the study's Series Instance UID and accession data to a secure, HIPAA-compliant inference queue. The AI service, typically containerized and GPU-accelerated, retrieves the anonymized images via DICOMweb WADO-RS, executes the analysis (e.g., for diabetic retinopathy grading, macular edema quantification, or glaucoma progression), and returns results as a DICOM Structured Report (SR) or a JSON payload containing measurements, confidence scores, and annotated overlays.

These AI-generated findings are then ingested back into Intelerad via its Clinical Data Repository API or by creating a new DICOM SR object linked to the original study. For the clinician, this manifests as an AI Findings Panel within the PowerReader viewer, showing quantified metrics (like retinal layer thickness) and detection flags. Critical results can be configured to trigger HL7 ADT messages to the EHR or create priority worklist flags. The architecture supports a human-in-the-loop review; the ophthalmologist can accept, modify, or reject AI suggestions, with all interactions logged to an audit trail for model performance tracking and regulatory compliance.

Rollout is phased, starting with a single AI algorithm (e.g., DR screening) in a pilot clinic. Governance is managed through Intelerad's role-based access controls (RBAC) to determine which users see AI results, coupled with a prompt library for consistent report drafting. The system is designed for scalability, allowing new containerized AI models to be added to the inference pipeline without modifying the core PACS integration, enabling health networks to evolve their ophthalmic AI portfolio over time. For related architectural patterns, see our guides on AI Integration for Vendor Neutral Archives (VNA) and AI Integration for Radiology Reporting Platforms.

A production integration for Intelerad Ophthalmology PACS is built on a secure, auditable pipeline. DICOM studies from modalities like OCT and fundus cameras are ingested via HL7 ORM/ORM messages or DICOM C-STORE to a secure staging area. AI inference runs in an isolated, HIPAA-compliant environment—often a private cloud or on-premises GPU cluster—where patient data is never persisted. Results are returned as DICOM Structured Reports (SR) or DICOM Secondary Capture objects, tagged with the original study's UIDs and stored back into the Intelerad VNA. This creates a permanent, traceable link between the AI output and the source images within the PACS audit trail.

Governance is enforced at multiple levels. Role-based access control (RBAC) within Intelerad determines which ophthalmologists or technicians can view AI findings, often restricting initial visibility to a "QA" or "Overread" worklist. AI-generated measurements or detection flags (e.g., "Referable Diabetic Retinopathy Suspected") are presented as non-destructive overlays or sidecar reports, requiring explicit clinician verification before being incorporated into the final signed report. A human-in-the-loop approval step is mandatory for critical findings, ensuring the AI acts as an assistive tool, not an autonomous decision-maker. All AI interactions, from study trigger to result review, are logged for performance monitoring and regulatory compliance.

A phased rollout minimizes disruption. Phase 1 (Pilot) typically targets a single, high-volume workflow like diabetic retinopathy screening, deploying AI for automated triage to prioritize urgent cases. AI results are visible only to a pilot group of clinicians in a separate worklist. Phase 2 (Expansion) integrates AI findings directly into the primary reading workflow for the pilot use case, using subtle visual cues (e.g., color-coded thumbnails) to indicate AI-prioritized studies. Phase 3 (Scale) adds additional AI models (e.g., for glaucoma progression, macular edema) and expands access across the department, with continuous feedback loops where radiologist corrections are used to retrain and improve model performance. This staged approach builds confidence, refines workflows, and delivers measurable impact—reducing time to diagnosis for urgent cases—before full-scale deployment.