Free resource for optometrists
AI for Eye Care
A living field guide for ODs: the current frontier AI models and where they actually fit in eye care, what the medical-technology companies in the space are shipping, and a no-noise way to stay up to date. No signup — just read, use, and share.
01 — The models
The current frontier models
As of August 2026, the leading general-purpose models are strongly multimodal, score near the top of medical-knowledge benchmarks, and offer long context with agentic tool use. Versions change monthly — treat the names below as a snapshot, not gospel.
Claude
Anthropic
Long-horizon research, agentic tool use, coding, image understanding, and professional knowledge work.
In your chair: Drafting and tightening notes, summarizing literature, and building practice tools and patient-facing copy.
Fable 5 · Sonnet 5 · Opus 4.8
Official model updateChatGPT
OpenAI
Multimodal analysis, web research, scientific reasoning, and long-running knowledge-work tasks.
In your chair: Clinical Q&A, analyzing a pasted document or chart, education drafts, and board/CE study.
Instant · Thinking · Pro
Official model updateGemini
Native multimodal and document understanding, fast production tiers, and Workspace integration.
In your chair: Parsing dense imaging reports and spreadsheets, and reasoning over images inside Google Workspace.
3.1 Pro · 3.6 Flash · 3.5 Flash-Lite
Official model updateLlama
Meta
An open-weight multimodal family that can be self-hosted for greater infrastructure and data control.
In your chair: For vendors and IT teams building HIPAA-controlled, on-prem tools — not for solo-practice use directly.
Open weight
Official model updateGrok
xAI
Coding, agentic tasks, and knowledge work, with live web and X context in the Grok product.
In your chair: Quick current-information lookups and general drafting when you want live web context.
Grok · SuperGrok · API
Official model updateA note on benchmarks
Top models now answer USMLE-style questions at ~94–97%, yet they stay far more reliable at text reasoning than at reading a fundus photo or OCT directly — and public benchmarks may overlap their training data. Treat high scores as directional, not a clinical guarantee.
Important: none of these general models is an FDA-cleared medical device. They are reasoning and productivity tools — not approved to diagnose from a fundus photo or OCT.
02 — In your chair
Where AI actually helps in eye care
The mature wins are in cutting administrative drag — documentation, coding, recall, communication — not autonomous diagnosis. Concrete, available-now uses, with the tools behind them.
Documentation & AI scribes
Ambient scribes listen to the exam and draft a structured note with suggested ICD-10/CPT in your EHR. Eye-care-native options include RevolutionEHR AI Scribe, Moyae, Doctora, and Barti; general tools include DAX Copilot, Abridge, and Suki. Highest-ROI use today — but every draft must be reviewed and signed.
Patient education
Draft and simplify handouts to a ~6th-grade reading level, translate them, or turn a clinical note into plain-language after-visit instructions. Always check medical accuracy, and keep patient identifiers out of consumer tools.
Triage & pre-screening
Summarize intake, answer common questions, and route patients before a visit. Useful for workflow — but never let a general chatbot gate red-flag symptoms like sudden vision loss, new flashes/floaters, or a painful red eye.
Imaging & OCT assistance
For actual clinical reads, use purpose-built, FDA-cleared tools — LumineticsCore, EyeArt, and AEYE for diabetic retinopathy; Altris AI, RetinAI, and the Notal OCT Analyzer for OCT — not general chatbots, which are not validated to interpret your patients' images.
Coding & billing
Code-suggestion engines read your documentation and propose ICD-10/CPT, and an LLM makes a useful coding sounding board. Note CPT 92229 for autonomous point-of-care retinal imaging, and that 2026 claims need laterality and severity to clear automated edits. You still own the claim.
Practice operations & marketing
Scheduling, recall texts, review responses, blog and social copy, newsletters, and SOPs. A low-risk entry point with little or no PHI involved — vendors report meaningful drops in no-shows and front-desk hours.
Board & CE study
LLMs make strong tutors: generate practice questions, explain concepts, and summarize papers. Verify against authoritative references — models are weakest on clinical optics and image-based reasoning, and can state wrong answers confidently.
Research & literature review
Clinician-grade, cited tools — OpenEvidence, Consensus, Elicit, scite — beat plain chatbots because they trace answers to sources. Always confirm the citations are real before you rely on them.
Before you deploy anything
Four guardrails that aren't optional
HIPAA, PHI & BAAs
Consumer chatbots (ChatGPT Free/Plus, the Claude.ai app, consumer Gemini) are not HIPAA-eligible. To use AI with patient data you need a signed Business Associate Agreement — available via the OpenAI API and ChatGPT Enterprise, the Anthropic API and Enterprise, Azure OpenAI, and Google Vertex AI / Workspace. Ambient-scribe vendors are Business Associates too: sign a BAA, update your Notice of Privacy Practices, and tell patients the visit is recorded.
Hallucination is real
General models can be confidently wrong and will fabricate citations. Reasoning models reduce this but do not eliminate it. Verify anything clinically significant before it touches a patient or a chart.
FDA status
No generative AI is FDA-cleared to diagnose. Only narrow, purpose-built tools — autonomous DR screening and home-OCT analysis — are cleared, each for one defined intended use. A general LLM describing an image is informal, not a clinical read.
Judgment & liability
The standard of care increasingly expects you to know when to override AI. Some states (California, Texas) now regulate AI in clinical communications and records. AI drafts; the OD reviews, signs, and remains responsible.
03 — Industry news
Eye-care medical technology, dated
A dated snapshot of what companies building AI and medtech for eye care have shipped, plus a few foundational clearances that still define the market. Every item links to a primary source; re-check it before you quote or buy.
Autonomous diabetic-retinopathy screening
The most mature clinical AI in eye care: FDA-cleared systems that return a point-of-care refer / no-refer result.
- Jan 2026
AEYE Health expanded Epic integration for its portable, fully autonomous AEYE-DS workflow; the handheld-camera indication itself received FDA clearance in April 2024.
AEYE Health - Sep 2025
Eyenuk secured the first broad national-health-system deployment of autonomous eye screening, beginning in South-Eastern Norway, a region serving 3.1 million people.
Eyenuk - Sep 2025
Banner Health deployed Digital Diagnostics' LumineticsCore across 24 centers and clinics in Phoenix and Tucson for autonomous diabetic-retinopathy exams.
Digital Diagnostics / Banner Health - Jun 2023
EyeArt added FDA-cleared support for the Topcon NW400, making it the first autonomous DR system cleared for cameras from multiple manufacturers.
Eyenuk
OCT & imaging AI
Imaging vendors are embedding AI into review workflows, and authorized home OCT is moving through a limited early rollout.
- Aug 2025
ZEISS earned a CE mark for CIRRUS PathFinder, which automatically flags abnormal macular OCT B-scans and enhances OCT-A image quality.
ZEISS - May 2025
Topcon acquired RetInSight, whose vendor-inclusive AI analyzes OCT images to detect and monitor geographic atrophy, wet AMD, diabetic macular edema, and retinal vein occlusion.
Topcon Healthcare - May 2024
The FDA granted De Novo authorization to Notal Vision's patient-operated SCANLY Home OCT and AI-based fluid-volume analysis for monitoring wet AMD between visits; rollout remains limited.
FDA / Notal Vision - Mar 2026
The FDA refreshed its public AI-enabled medical-device list, giving clinicians a better primary-source starting point for checking authorization status and intended use.
U.S. FDA
Smart eyewear & the optical channel
Consumer AI glasses aren't medical devices — but they're dispensed through optical retail and they drive patient questions.
- Sep 2025
EssilorLuxottica and Meta launched the Meta Ray-Ban Display ($799) — the first heads-up-display smart glasses with a neural wristband — alongside an expanded AI-glasses line.
Meta - Jun 2026
EssilorLuxottica and Meta launched a $299 Meta Glasses collection with three prescription-compatible styles, expanding AI eyewear into more optical retail channels.
EssilorLuxottica - May 2026
Google previewed Gemini-powered Android XR eyewear from Warby Parker and Gentle Monster, with audio-first models planned for fall 2026.
Google - Mar 2026
Ray-Ban Meta Optics added optical-first, prescription-focused frames while expanding Meta AI features such as live translation and message summarization.
EssilorLuxottica
04 — Stay current
A system for keeping up
You can't read everything, and you don't need to. Pick a couple of items from each list, then follow the simple routine at the bottom.
Newsletters
- Glance by Eyes On Eyecare
weekly, free
- Optometry Times
clinical + an AI beat
- Review of Optometry
Optometric Physician digest
- Healio — Primary Care Optometry News
news + CME
- The Batch — DeepLearning.AI
best weekly for 'why it matters'
- The Rundown AI
a fast daily for general AI
Journals
- Translational Vision Science & Technology
ARVO, open access
- Ophthalmology Science
AAO, open access
- JAMA Ophthalmology — “Eye on AI”
dedicated AI theme
- Optometry & Vision Science
flagship OD research journal
- NEJM AI
evidence standards for clinical AI
- PubMed alert
save a search; let papers come to you
Podcasts
- From Paper to Clinic
Optometry Times — start with the AI episode
- Defocus Media
OD network, regular tech coverage
- Hard Fork
NYT — accessible big-picture AI
- Practical AI
applied AI for non-engineers
Upcoming meetings to watch for AI tracks
- Academy 2026 (AAOpt)
Sep 30 – Oct 3, 2026
Anaheim, CA
- Vision Expo 2027
Mar 10 – 13, 2027
Las Vegas, NV
- ARVO 2027
May 2 – 6, 2027
San Diego, CA — vision research frontier
- AOA Optometry's Meeting 2027
Jun 16 – 19, 2027
Boston, MA
- Academy 2027 (AAOpt)
Nov 10 – 13, 2027
San Antonio, TX
Earn AI-focused CE through CEwire, AOA+ education, and Eyes On Eyecare — and confirm any course is COPE-approved and accepted by your state board via the ARBO/COPE database.
People & communities to follow
Michael Abràmoff, MD, PhD
Digital Diagnostics — pioneer of autonomous AI in eye care
Pearse Keane, MD
Moorfields / DeepMind retinal-AI research
Daniel Ting, MD, PhD
Singapore National Eye Centre — DR deep-learning screening
A simple routine
Daily · 5 min
Skim one general AI newsletter. Just note the headlines that touch medicine or imaging.
Weekly · 30–45 min
Read Glance for eye-care industry news, one deep-dive from The Batch, and listen to a podcast on the commute.
Monthly · ~1 hr
Check your PubMed alert or a journal's table of contents, and try one AI tool in a low-risk workflow.
Quarterly / yearly
Attend one conference's AI track, earn AI-focused CE, and re-check which cleared tools are reimbursable and in scope for your state.
About the editor
Eugene Huang, O.D.
Optometrist · applied-AI builder · founder of PreTectal
Eugene Huang, O.D., works at the intersection of optometry and technology, with a focus on making artificial intelligence useful to optometrists and eye-care teams.
Through PreTectal, he evaluates frontier AI models, translates them into practical clinical and operational workflows, and builds products including OptoConnections and WhatsTheDx. This guide combines optometric training with hands-on AI product development so ODs can separate durable tools from hype.
Eugene Huang on LinkedInApplying AI in your practice?
This guide is free and updated as the field moves. If you want a hand choosing tools, fitting them to your workflow, or building something for your practice, PreTectal is happy to talk.