Radiology Service — AI Integration
Status: populated Owner: TBD Last updated: 2026-04-18 Companion: Service Template · 03 platform-services · 02 DDD
1. Current AI Usage
The radiology-service does not directly invoke any AI/ML model in its baseline. No vendor AI SDKs are used. Per platform policy, any future assistive AI features must route through ai-gateway-service via Kong /v1/ai/*.
2. Planned / Future AI Features (Tier B — Assistive)
| Feature | Purpose | Prompt template | HITL | Status |
|---|---|---|---|---|
| Radiology report summary | Auto-draft a plain-language summary for referring clinician | rad.report.summarize.v1 | Yes — radiologist must review before filing | Not implemented |
| Critical finding detection assist | Highlight potential critical patterns for radiologist review | rad.finding.critical-assist.v1 | Yes — radiologist makes final determination | Not implemented |
| Structured report extraction | Extract structured data fields from narrative report text | rad.report.extract-structured.v1 | Yes | Not implemented |
3. Moderation Policy
- AI outputs are suggestions only; never auto-filed.
- PHI in prompts is scoped to minimum-necessary fields (report text only, not full patient record).
AIProvenancemetadata recorded when radiologist accepts an AI-assisted draft.
4. Rationale
Radiology AI (CADe/CADx) is typically embedded in the PACS or modality vendor. The EHR-level AI assist is supplementary workflow tooling, not a diagnostic AI. Primary diagnostic AI is out of scope for the platform baseline.