AI Integration for GE Surgery Imaging

AI integration for GE surgery imaging focuses on three primary surfaces: the acquisition console, the advanced visualization workstation (AW Suite), and the perioperative documentation system. At the console, AI models can connect via GE's Edison AI Platform APIs to analyze live fluoroscopy or 3D rotational angiography feeds for real-time dose optimization and image quality enhancement, providing instant feedback to the technologist. Within the AW Suite, AI-powered segmentation tools can be embedded to automatically delineate anatomical structures (e.g., spinal pedicles, tumor margins, vascular anatomy) from intraoperative CT or cone-beam CT scans, overlaying 3D models directly onto the surgeon's navigation display.

The high-value workflow is automated procedural documentation. By integrating with the surgery imaging system's DICOM metadata and surgical information system (via HL7), an AI agent can listen for study completion events. It then triggers a multi-step process: 1) Analyze the final intraoperative images to identify key steps (e.g., instrument final position, implant placement, contrast injection phases), 2) Draft a structured procedural note summarizing findings, measurements, and dose reports, and 3) Route the draft to the surgeon's mobile device or EHR inbox for review and sign-off via a secure API. This turns a manual, post-case documentation task into a same-room activity, capturing details while the procedure is fresh.

A production implementation is typically wired through a secure, on-premise or hybrid cloud gateway. DICOM images are sent from the GE modality to a dedicated inference server (often GPU-accelerated) via DICOM C-STORE. The AI results—structured as DICOM Structured Reports (SR) or JSON payloads—are sent back to the PACS for archival and to the AW Suite for display. A parallel workflow engine, perhaps built on n8n or a similar orchestrator, manages the documentation drafting and approval routing, ensuring all actions are logged for audit trails. Rollout should start in a single OR suite, focusing on a repeatable procedure like spinal fusion or embolization, to validate the AI's accuracy and the surgeon's trust in the automated outputs before enterprise scaling.

Governance is critical. AI suggestions in the OR must be clearly presented as assistive tools, not autonomous decisions. Integration should include a mandatory human review step for all documentation drafts and a simple feedback mechanism (e.g., a 'correct/incorrect' button) within the surgeon's UI to continuously improve model performance. Furthermore, AI-driven dose alerts must be configured with institution-specific thresholds and should not interrupt the surgical flow without explicit, configurable rules. This controlled, workflow-aware approach ensures AI augments the surgical team's expertise without introducing new risks or delays.

Integration begins at the modality. For GE C-arm and O-arm systems, AI models connect via the GE Edison AI Platform APIs or a secure DICOM router. Intraoperative images (fluoroscopy, 3D scans) are streamed to a local inference server or cloud endpoint. The architecture must support low-latency processing—often sub-second for real-time guidance—and maintain DICOM compliance for image integrity and metadata preservation. Key data objects include the Series Instance UID, Study Description, and Acquisition Context to trigger the correct AI algorithm (e.g., dose optimization vs. instrument detection).

The processed AI results—such as enhanced images, dose maps, or instrument overlays—are returned to the surgical workflow through two primary paths: 1) Direct injection back to the PACS as a new DICOM series (e.g., Softcopy Presentation State or Segmentation objects) for review on the surgical monitor, or 2) Via a REST API to the GE Surgery Guidance application or a custom surgical dashboard for real-time visualization. For automated documentation, AI-generated observations (e.g., "peak skin dose: X mGy", "critical structure proximity: Y mm") are formatted into HL7 FHIR or DICOM Structured Report (SR) and pushed to the EHR or surgical reporting module via an integration engine.

Rollout requires a phased, procedure-specific approach. Start with a single AI application (e.g., low-dose fluoroscopy enhancement) in a controlled OR, integrating with the GE Allia™ IGS platform or Advantage Workstation. Governance is critical: establish a QA loop where AI outputs are logged and compared against physicist and surgeon feedback. Implement RBAC to control which AI tools are available per user role (e.g., surgeon vs. technologist). All data flows must be encrypted in transit, and AI models should be validated against the hospital's specific device models and protocols to ensure clinical accuracy and safety.

A production integration for GE surgery imaging (C-arm, O-arm) must enforce strict data governance from the start. This means implementing a zero-trust architecture where AI inference runs in a secure, isolated environment—either on-premises or in a HIPAA-compliant cloud enclave. DICOM images and associated patient metadata are never persisted by the AI service; they are streamed via secure DICOMweb or REST APIs from the PACS/VNA (e.g., GE Centricity, HealthCloud) for transient analysis. All AI-generated outputs, such as dose optimization suggestions or image quality scores, are returned as DICOM Structured Reports (SR) or HL7 messages, creating a permanent, auditable trail within the existing imaging record. Role-based access control (RBAC) from the PACS or hospital AD must govern which surgeons and technologists can view or act on AI insights.

A phased rollout is critical for clinical validation and workflow integration. We recommend a three-phase approach: 1) Silent Mode: AI processes live imaging data in the background, generating logs and predictions without any display in the OR. This builds a performance baseline and identifies edge cases. 2) Assistive Mode: Non-critical, high-confidence AI suggestions (e.g., 'Low Dose Alert' or 'Collimator Adjustment Recommended') are displayed as non-interruptive notifications in the imaging workstation UI. Staff can acknowledge or ignore them, building trust and gathering feedback. 3) Integrated Mode: After validation and protocol refinement, key AI functions are embedded into the procedural workflow—for example, automated fluoro-loop capture for documentation or real-time overlay guidance for instrument positioning. Each phase requires defined success metrics, such as reduction in repeat acquisitions or time saved per procedure.

Ongoing governance is managed through a centralized LLMOps or MLOps dashboard that monitors model drift, inference latency, and user engagement. A clinical champion—often a lead surgeon or imaging director—should oversee a monthly review of AI performance logs and adverse event reports (even false positives). This ensures the AI remains aligned with evolving clinical protocols and hardware updates. For a sustainable program, the integration should be designed to allow for the safe swapping of AI models (e.g., from a third-party vendor for fracture detection to an internally developed model for tissue characterization) without disrupting the core imaging workflow, using containerized services and a well-defined API contract.