# EYRIX — AI Decision Support for Ophthalmology

> EYRIX is a clinical-AI platform that turns any OCT or fundus image into a structured, specialist-grade retinal report in seconds. Built by a medical-imaging ML team, it combines pixel-level retinal-layer segmentation, OCT disease classification (DME, CNV, Drusen, AMD, CSR, Macular Hole, VID, Normal), OCT biomarker localisation, and 5-stage Diabetic Retinopathy grading from fundus photographs. The platform is scanner-agnostic, delivered as a REST API, and engineered for clinical safety — every result carries a confidence level, borderline cases are routed for specialist review, and uncertainty is surfaced rather than hidden. EYRIX is part of the LivConnect health ecosystem.

EYRIX is built for ophthalmology clinics, eye-care hospital networks, tele-ophthalmology providers, government screening programmes, and OCT device manufacturers who need consistent, explainable AI-assisted reads at scale.

## Core capabilities

- **Multi-layer retinal segmentation** — pixel-level identification of 10 distinct OCT structures: RNFL, GCL+IPL, INL, OPL, ONL, IS/OS (ellipsoid zone), RPE, Choroid, plus intraretinal and subretinal Fluid.
- **OCT disease classification** — dual-model router covering CNV, DME, Drusen, AMD, CSR, Macular Hole, VID, and Normal, with deterministic conflict resolution and an explicit `review_required` flag for low-confidence cases.
- **OCT biomarker detection** — two-stage RetinaNet + DenseNet-169 verifier that localises soft drusen, hard drusen, reticular drusen, soft-drusen PED, photoreceptor-layer disruption, choroidal folds, and geographic atrophy with confidence scores and bounding boxes.
- **Diabetic Retinopathy grading (fundus)** — two-stage DenseNet-201 pipeline (binary screening + CORAL ordinal grading) producing five-grade severity: No DR, Mild, Moderate, Severe, Proliferative DR. Optional ensemble path uses EfficientNet-B7/B4, ConvNeXt and ResNet-50.
- **Volume-level intelligence** — the segmentation model runs on every B-scan in an OCT volume, not just the central slice, and surfaces peripheral findings the classifier would miss.
- **Visual explainability** — Grad-CAM overlays and colour-coded layer maps showing exactly which regions drove the AI's finding.
- **Clinical safety architecture** — automatic image quality gating, per-class confidence thresholds, entropy-based abstention, anatomical sanity checks (layer ordering, thickness clamps), and CSR-specific dome reconstruction.

## What EYRIX is and is not

- EYRIX is a **screening aid and decision-support layer**, not a standalone diagnostic device. Every output is intended to support — not replace — clinical judgement by a qualified ophthalmologist.
- EYRIX is currently for **research and investigational use only**. It has not been cleared or approved by FDA, CE, or CDSCO for clinical diagnostic use.
- EYRIX never recommends specific treatments (anti-VEGF, laser, surgery or otherwise). It reports findings and confidence; the clinician decides.

## Technical facts

- **Delivery:** REST API (FastAPI backend), deployable on cloud or on-prem (Docker). Authentication via short-lived `X-EYRIX-HEADER` tokens.
- **Inputs:** OCT B-scans (single or full volume, SD-OCT or SS-OCT) in DICOM/TIFF/PNG/JPEG; colour fundus photographs.
- **Outputs:** structured JSON with predicted class + probabilities, per-layer thickness arrays, fluid coverage percentage, biomarker bounding boxes with confidence, base64-encoded colour overlay JPEG, and an explicit `review_required` flag with a human-readable reason.
- **Models:** U-Net with ResNet-34 encoder (segmentation); DenseNet-169 (OCT classification, OCT Model A); 5-class expansion model (OCT Model B); two-stage DenseNet-201 (DR grading); RetinaNet + DenseNet-169 (biomarkers).
- **Latency:** sub-second per B-scan on a single GPU; designed to return a structured AI assessment per scan in under a minute end-to-end.
- **Scanner support:** designed to be device-agnostic; current models trained primarily on SD-OCT data and being validated across Spectralis, Cirrus, Optovue, and Topcon.

## Conditions detected

- Diabetic Macular Edema (DME)
- Choroidal Neovascularization (CNV) / wet AMD
- Drusen and dry AMD spectrum
- Central Serous Retinopathy / CSCR
- Macular Hole (MH)
- Vitreomacular Interface Disease (VID)
- Geographic Atrophy
- Diabetic Retinopathy — 5 stages (No DR, Mild, Moderate, Severe, Proliferative DR)

## Who it is for

- Ophthalmology clinics and retina specialists who want a consistent second-reader.
- Eye-care hospital chains standardising diagnostic quality across decentralised centres.
- Tele-ophthalmology and remote-screening platforms needing an API-driven backend.
- Government and NGO mass-screening programmes (Ayushman Bharat, NPCB-VI, World Diabetes Foundation).
- OCT and ophthalmic device OEMs bundling an AI software layer with their hardware.
- Optical-retail chains and corporate diagnostic labs adding premium retinal-screening services.

## Markdown shadows (clean copies for LLMs)

- [Home](https://eyrix.ai/index.html.md): Overview of the EYRIX platform — hero, value props, capabilities, clinical-safety architecture.
- [About](https://eyrix.ai/about.md): The team, research foundation, and product philosophy behind EYRIX.
- [Capabilities](https://eyrix.ai/capabilities.md): Detailed breakdown of OCT layer segmentation, fluid detection, biomarkers, and DR grading.
- [Conditions](https://eyrix.ai/conditions.md): The retinal conditions EYRIX detects, with clinical context for each.
- [Workflow](https://eyrix.ai/workflow.md): How EYRIX integrates into an imaging workflow — upload, quality check, AI analysis, structured report.
- [Clinical Safety](https://eyrix.ai/clinical-safety.md): The safety architecture — quality gating, confidence thresholds, abstention rules, anatomical sanity checks.
- [API](https://eyrix.ai/api.md): REST API surface, authentication, request/response shapes, deployment models.
- [FAQ](https://eyrix.ai/faq.md): Common questions about EYRIX — regulatory status, accuracy, scanner support, pricing model, data handling.

## Contact

- General: hello@eyrix.ai
- Request a demo: demo@eyrix.ai
- Partnerships (hospitals, OEMs, screening programmes): partners@eyrix.ai
- Research / data partnerships (institutes): research@eyrix.ai

## Optional

- [Sitemap](https://eyrix.ai/sitemap.xml)
- [Robots](https://eyrix.ai/robots.txt)
- [Parent ecosystem — LivConnect](https://www.livconnect.ai/)