AI in medicine, explained for everyoneGeneral information only, not medical advice · Our standards
How Far

Eyes · Diagnosis and imaging

AI eye screening in Tanzania: more people reached the eye clinic

Evidence: Moderate
How it was checked
AI-assisted source check. Not clinically reviewed. ()
Last updated
Next review
Page ID
TR-017

Our editorial standards · 7 references · Corrections

How far from you?

The track: 1 Lab → 2 Tested in people → 3 Approved somewhere → 4 Approved in this country → 5 Routine care in this country. Each region sits at the furthest stop backed by a record.

RegionHow farStatus and record
United States Stop 2 of 5: Tested in peopleRegister not searched yet
European Union Stop 2 of 5: Tested in peopleRegister not searched yet
United Kingdom Stop 2 of 5: Tested in peopleRegister not searched yet
Australia Stop 2 of 5: Tested in peopleRegister not searched yet

Tanzania: Register not searched yet. The authors wrote in 2024 that Tanzania had no regulatory framework for approving AI-assisted devices. The Ministry has committed to roll-out (see c6); no TMDA record found.

Evidence strength
Moderate. An individually randomised trial of 2,378 adults at 24 hospitals (Lancet Global Health 2026). Short (8-week outcome), measures clinic attendance rather than sight saved, one region, and can't separate the AI from the extra counselling.
Next milestone
No date announced.

General information only, not medical advice. Talk to your doctor about your situation. Read the disclaimer.

In a randomised trial of 2,378 people with diabetes at 24 Tanzanian hospitals, AI eye screening with an on-the-spot result and face-to-face counselling roughly doubled clinic attendance among the 904 who needed referral (35.9% vs 18.2% within 8 weeks). It measured attendance, not treatment or sight saved.

How far away is this?

Evidence: Moderate

It's a completed randomised trial, but the main outcome is clinic attendance, not treatment received or sight saved. Even with AI, nearly two-thirds (64%) of people who needed referral didn't attend within 8 weeks, and the trial can't separate the AI from the extra counselling.

Where is it approved?

RegionRegulatorStatusNotes
United StatesFDARegister not searched yet Source previously cited:[7]
European UnionEUDAMEDRegister not searched yet Source previously cited:[7]
United KingdomMHRARegister not searched yet Source previously cited:[7]
AustraliaTGARegister not searched yet Source previously cited:[7]
TanzaniaMinistry of Health / TMDARegister not searched yetThe authors wrote in 2024 that Tanzania had no regulatory framework for approving AI-assisted devices. The Ministry has committed to roll-out (see c6); no TMDA record found.[6]

The trial used SELENA+ by EyRIS. A 2025 peer-reviewed review lists it as CE marked and on Australia's ARTG (both Class IIa); we couldn't confirm either in the official registers. We found no FDA clearance. This status is for the software in general, not for the screening pathway tested in this trial. "Register not searched yet" means we haven't searched that regulator's register; it doesn't mean the product is or isn't available. Approval in one country isn't approval in another.

What is it?

Photos of the back of the eye were read straight away by AI software (SELENA+), so the patient got their result and face-to-face advice before leaving. In the usual service in this study, the photos were graded weeks later and results were given by phone or text message.

What's the evidence?

  1. Two-arm, parallel-group, individually randomised (1:1) controlled trial at 24 hospitals in Kilimanjaro and Arusha, Tanzania. 2,378 adults with diabetes were enrolled between March 2023 and August 2024: 1,192 to AI-supported screening and 1,186 to standard care.[2]

    Evidence stage: In clinical trials · #c1

  2. Primary outcome: attendance at the central eye clinic within 8 weeks among people that masked UK-certified graders judged to have truly referable retinopathy (904 of 2,378). It was 157/437 (35.9%) with AI vs 85/467 (18.2%) with standard care: a difference of 17.7 percentage points (95% CI 12.0–23.4), odds ratio 2.52 (95% CI 1.86–3.42).[2]

    Evidence stage: In clinical trials · #c2

  3. According to the university's news release, people who did attend got there sooner after AI screening: an average of 10.9 days vs 36.9 days. This compares only those who attended, not everyone randomised, and we haven't checked it against the paper.[4]

    Evidence stage: In clinical trials · #c3

  4. The AI group got two things at once: an immediate AI result and face-to-face counselling (the standard group got their result weeks later by phone or text message). So the trial can't show how much of the effect came from the AI itself. The trial is single-masked: participants and screening staff couldn't be blinded, but the graders who set the reference standard and the staff recording attendance were masked.[5]

    Evidence stage: In clinical trials · #c4

  5. Funded by the British Council for the Prevention of Blindness, Christian Blind Mission and the Sir Halley Stewart Trust. Some authors declare industry fees (e.g. Roche, Novartis, Heidelberg Engineering); the others declare none. No author declares a link to EyRIS, which makes the AI software. The protocol thanks EyRIS for providing the software.[2]

    Evidence stage: In clinical trials · #c5

  6. Tanzania's Ministry of Health, a collaborator on the study, has committed to adding AI diagnostics to diabetic eye screening in several regions.[4]

    Evidence stage: Announced only · #c6

  7. The trial registry names the AI device as SELENA+, and the published protocol identifies its maker as EyRIS (Singapore). The registry also lists secondary outcomes, including how many people started treatment, the AI's accuracy against expert graders, and cost-effectiveness. These aren't in the abstract and we haven't seen them.[3]

    Evidence stage: In clinical trials · #c7

What does it mean for me?

If you have diabetes, regular eye screening is generally recommended; ask your doctor or diabetes team how often you need it. This trial suggests that getting your result and advice on the spot can help more people get to an eye clinic, which could matter wherever specialists are far away. It doesn't tell you which screening service or product to use.

What could I ask my doctor?

  • When will I get my eye screening result?
  • If I need referral, how and when do I get the appointment?
  • Is any AI used to read my eye photos, and does a person check the results?

These are general starting points, not a recommendation to choose any particular test or treatment.

Change history

  • : First published.
  • : Review label corrected: this page was not clinically reviewed. Earlier labels saying 'independent fact-check' and 'clinical review' were wrong; checks so far were AI-assisted source checks only.

General information only, not medical advice. Talk to your doctor about your situation. AI-assisted source check. Not clinically reviewed.

References

  1. Cleland et al., Lancet Global Health (2026), DOI. doi.org (2026). Peer-reviewed randomised trial. https://doi.org/10.1016/j.langlo.2026.104062
  2. PubMed abstract and declaration of interests (PMID 42863478). pubmed.ncbi.nlm.nih.gov. Abstract. https://pubmed.ncbi.nlm.nih.gov/42863478/
  3. ISRCTN18317152. isrctn.com. Trial registry. https://www.isrctn.com/ISRCTN18317152
  4. LSHTM ICEH news release. iceh.lshtm.ac.uk. University news. https://iceh.lshtm.ac.uk/people-with-diabetic-retinopathy-nearly-twice-as-likely-to-reach-follow-up-with-ai-enabled-screening/
  5. Cleland et al., trial protocol, BMJ Open (2024). doi.org (2024). Peer-reviewed protocol. https://doi.org/10.1136/bmjopen-2023-075055
  6. Cleland et al., implementing AI screening in Tanzania, Trans R Soc Trop Med Hyg (2024). doi.org (2024). Peer-reviewed study. https://doi.org/10.1093/trstmh/trae132
  7. Scoping review of ophthalmic AI medical devices, npj Digital Medicine (2025), Table 1. nature.com (2025). Peer-reviewed review (regulatory status). https://www.nature.com/articles/s41746-025-01726-8/tables/1

Publisher shown as the website hosting each source. Dates are given where the source label includes them.