A nurse in a Beijing community clinic slides a patient into a small chair, points a camera the size of a desk lamp at one eye, and clicks. Sixty seconds later a report lands on screen: signs of diabetic retinopathy, graded and flagged. No ophthalmologist is in the room. The "doctor" is an algorithm trained on millions of labeled eye images, and it fits, camera and all, into a single backpack.
This is the quiet front line of China's medical-AI deployment — and one company, Airdoc (鹰瞳科技), has turned it into the country's largest installed base of retinal screening points.
Key takeaways
- Airdoc's retinal AI now runs at 8,599 active service points (up 9.1% year on year), including 478 hospitals and 408 physical-exam centers (2025 annual report, filed with HKEX).
- Its core algorithm platform, Airdoc-AIFUNDUS (2.0), received China's NMPA Class III medical-device registration in April 2025, covering diabetic retinopathy and retinal vein occlusion.
- In 2025 the system performed 488,000 detection sessions (+21.6%) and, across its full screened population, flagged 37,210 severe-positive cases — early warnings for conditions that often go unnoticed until vision is lost.
- A portable, offline-capable fundus camera lets the same AI reach rural clinics with no specialist and no internet.
Why the eye is the entry point
The retina is the only place in the body where blood vessels and nerve tissue can be viewed directly without surgery. Changes there track not just eye disease but diabetes, hypertension, and cardiovascular risk. The catch: reading a fundus photo well takes years of specialist training, and China has far too few ophthalmologists for its population.
Airdoc's answer was to feed a model nearly 4 million fundus images, each repeatedly annotated by teams of tier-3 hospital experts, so the algorithm could learn the labeling habits and lesion standards of experienced doctors. The company says its diabetic-retinopathy module matched physician accuracy in multi-center trials and earned China's first Class III registration for an ophthalmic AI software product (the earlier DR product, registration number 国械注准20203210686, was approved in August 2020).
From hospital rack to backpack
The first models ran on large, costly hospital devices — powerful but stuck inside big hospitals. The engineering pivot was optical: by redesigning the lens system and pushing alignment steps into software, the team says it cut hardware cost by roughly an order of magnitude and built a portable camera small enough to carry in a backpack, with long battery life and local caching that works with no network.
That portability is the deployment story. The Charity Federation has taken the cameras into resource-poor regions for chronic-disease early screening. In Beijing, a co-developed AI fundus model from Tongren Hospital and Airdoc has landed in more than 100 grassroots clinics (reported by Worker's Daily in August 2026). In far-western Xinjiang, a county hospital doctor quoted by health media said the system let local patients "enjoy the same level of diagnosis as experts in Shanghai and Beijing" from 2020 onward.
The numbers behind the rollout
Airdoc's 2025 annual results, filed with the Hong Kong Exchange, give a rare quantified view of a medical-AI business at scale:
- Active service points: 8,599, up 9.1% from 7,883.
- Hospital coverage: 478 active outlets, up 14.6%.
- Detection sessions: 488,000, up 21.6%.
- Physical-exam centers deployed: 408; optometry-chain outlets reached 3,569 (+26.2%).
- Full-year screened population: 7.12 million user-visits, of which 37,210 were flagged as severe-positive (+4.4%).
- Retina-AI product revenue: RMB 120.8 million (≈ US$16.9M), up 7.2%.
- Total 2025 revenue: RMB 173 million (≈ US$24.2M), up 10.8%; net loss narrowed 90.2%.
The company listed on HKEX in November 2021 (stock code 02251.HK) as China's "first medical-AI stock" (医疗AI第一股). Its AI products have also reached overseas markets including the EU, Singapore, Thailand, Malaysia, and South Africa, with the company citing screenings across 50-plus countries.
Honest limitations
- Figures are company-reported. The 8,599 points, 488,000 sessions, and 37,210 severe cases come from Airdoc's own HKEX filings and earnings materials, not an independent clinical audit.
- "Severe-positive" is Airdoc's internal category, not a standardized clinical endpoint; it signals referral-worthy findings, not confirmed diagnoses.
- The global "30 million screenings / 50 countries" claim is the company's cumulative marketing figure and cannot be independently verified here.
- NMPA Class III registration confirms the device met regulatory safety and efficacy thresholds for a specified intended use; it is not an endorsement of outcomes in every deployment setting.
- Airdoc's retinal AI is a screening and clinician-support tool, not a replacement for examination by an ophthalmologist.
What readers can do now
- If you are in China and over 40, diabetic, or hypertensive, ask your clinic or annual physical whether a retinal AI screen is available — it is a low-cost, one-minute add-on at many exam centers and community sites.
- Treat a "positive" AI flag as a referral signal, not a verdict. Follow up with an eye specialist; the value is catching risk early, not self-diagnosing.
- For health-system watchers, track NMPA clearances by intended condition — the regulatory boundary between "health-risk assessment" and "diagnostic aid" determines where these cameras may legally operate.