Platform Overview
The AI-native RIS built for how radiologists actually work.
Flexible integrations with frontier LLMs and speech engines, ACR-scored US reporting, a self-improving prompt loop, and PACS integration — all in one clinical-grade platform.
Sign In to FLAIRSee it in action
Watch the demos
Platform overview
FLAIR + Fujifilm Synapse PACS integration
Voice dictation → AI-powered report generation
What makes FLAIR different
Five core capabilities that define FLAIR
Anti-Anchoring Guard on Every LLM Call
A hard-coded medical safety instruction is appended to every LLM prompt, regardless of which model or exam type is selected. It prohibits the model from restating the clinical indication as an imaging finding and requires explicit negation of unconfirmed conditions — e.g., "No CT evidence of acute intracranial injury" rather than echoing "rule out intracranial injury." This directly addresses diagnostic anchoring, a documented hallucination risk specific to AI in radiology. The guard is not configurable and cannot be overridden by prompt changes.
Six Speech Engines and a Custom Radiology Whisper Model
FLAIR routes across OpenAI Whisper, Google Medical Dictation (the medical-grade medical_dictation model at 16 kHz LINEAR16), Azure Cognitive Services, Deepgram Nova-3 Medical, Together.ai, and a local WebSocket Faster Whisper server — GPU-accelerated on NVIDIA hardware, CPU int8 otherwise. The local server runs a custom fine-tuned whisper-radiology-tiny model, domain-primed with DICOM, PACS, MRI, CT, and modality-specific vocabulary. Every call is logged by provider, duration, bytes, and error type. The admin dictation dashboard shows per-provider success rates and top 5 error types over 7/30/90-day windows — so you know which engine to lean on.
Prompts That Learn From Every Report
Every submitted report carries a 0–1 edit similarity score computed against the AI draft. Reports with low scores (heavily rewritten) surface in the quality dashboard, sortable by modality and exam type. One click sends the AI draft, the final report, and the current prompt to Gemini — which returns structured suggested_prompt_changes and a recommended_model. Prompts are first-class database objects: create, version, test in-browser with a live test harness, bind to specific exam types, and assign a model per prompt — no code changes, no redeployment.
One Keystroke from Synapse to the Report
A first-party Chrome/Edge extension for Fujifilm Synapse reads MRN, patient name, and exam type directly from the PACS web UI. Press Ctrl+Shift+F: the extension fuzzy-matches against the live AVR worklist — handling B-prefix, digit-only, and substring MRN variants — falls back to the FLAIR database, and redirects straight to the correct report-create screen with patient context pre-loaded. No manual lookup. No copy-paste. No tab switching.
US PDF to Structured Report with ACR Scoring
Upload an ultrasound order or prior report PDF and FLAIR parses the clinical context, applies the appropriate ACR scoring system — ACR TI-RADS for thyroid nodules, BI-RADS for breast, PI-RADS for prostate, LI-RADS for liver — and generates a pre-filled, structured report with risk stratification and follow-up recommendations baked in. The radiologist opens the report editor to a draft that already reflects the clinical indication, the scoring category, and the standard management recommendation for that score. No manual lookup of ACR lexicons, no copy-paste from order PDFs.
Full Platform
Everything in the platform
From worklist to signed report — and the infrastructure behind it.
Three LLM Providers
Google Gemini (default), OpenAI GPT-4o / o1 / o4-mini, and Amazon Bedrock (Claude Opus 4.6 through Haiku). Switchable per exam type, no redeploy.
Smart Templates
Auto-inject exam name, procedure, and patient history from the order. Three-level fallback cascade: exact exam type → modality + body part + contrast → modality + body part.
Quick Findings Library
Per-user snippet library keyed to exam type. Three-layer fallback so relevant findings appear even for uncommon exam variants. Insert a full structured finding in one click.
Prompt Management
Full CRUD on AI prompts. Bind to specific exam types and report styles. Model assigned per prompt. Test in-browser with live harness before publishing.
Report Quality Dashboard
Similarity-scored per report. Sort by most-edited to find which exam types need prompt tuning. One-click Gemini analysis returns actionable prompt changes.
Report Styles
User-selectable structured (organ-system bullets) or free-text narrative output. Style preference respected at prompt selection time.
Generation Audit Trail
Full system + user messages logged to AiGenerationLog on every call. Reviewable per report for compliance and debugging.
Multi-Study Reports
Tabbed interface for reports covering multiple studies in one session. Independent template selection per study tab.
US PDF → ACR Scoring
Upload ultrasound order PDF — FLAIR extracts clinical context, applies TI-RADS, BI-RADS, PI-RADS, or LI-RADS, and pre-fills a structured report with risk category and follow-up recommendation.
Six STT Backends
OpenAI Whisper, Google Medical Dictation, Google General, Azure Cognitive Services, Deepgram Nova-3 Medical, Together.ai. All selectable per deployment.
Local Offline Whisper
WebSocket Faster Whisper server runs on-premises, GPU float16 or CPU int8. Sliding-window streaming with VAD silence detection and a custom radiology-tuned model.
PowerMic III, No Dragon
Nuance dictation hardware works out of the box with open-source key bindings triggering Google Medical STT. No Dragon Medical license required.
Dictation Analytics
Provider breakdown, daily volume chart, success rate, and top 5 error types per engine — over 7, 30, or 90-day windows. Admin-only dashboard.
Inpatient & Outpatient Worklists
Proxies AVR teleradiology worklist HTML, merges FLAIR-only entries, and applies per-radiologist hospital visibility filters.
Four Role Types
Radiologist, Technologist, Admin, System Admin — each with a dedicated dashboard and fine-grained hospital privilege control.
Fax Automation
Fax queue, hospital phone/fax number lookup, and report faxing directly from the submission flow. No separate fax client needed.
Productivity Reports
Personal and group reading volume by modality, hospital, and date range. Timezone-aware. Visualized with Chart.js breakdown charts.
Natural Language DB Queries
Admins query the live database in plain English. Gemini converts to SQL, executes with user-context filtering. Neon and Aiven MCP both supported.
Monthly Reports Dashboard
Hospital-scoped monthly report counts, year/month selectable. JWT-scoped: non-admins see only their hospital's data.
Ready to see FLAIR in your workflow?
Sign in and start generating AI-assisted reports today.
Sign In to FLAIR