Between February and August 2024, Lions Outback Vision took a retinal camera in a van to nine communities in the Pilbara. A research officer with no formal healthcare qualification took the photographs. Two AI systems, Thirona RetCAD and EyRIS SELENA+, read them for diabetic retinopathy, and patients were told the result on the spot. A remote clinician supervised the screening and saw patients with referable disease by telehealth. Of the photographs, 10.3 percent showed referable retinopathy and 8.4 percent could not be graded. Eleven times as many people were screened as by local primary care cameras in the same months of 2023 [1].
Lions Outback Vision has also tested Google's Automated Retinal Disease Assessment (ARDA). By April 2026 a spin-out, Ninox Vision, had supplied it to seven general practices and Aboriginal medical services in Western Australia and Darwin [2]. In October 2025 a Diabetes WA article said that "the regulatory steps for implementation of new software as a medical device need to be undertaken" [3]. Neither source says how the software is regulated today.
The van is one of five points at which AI now comes into an eye examination.
- Before the clinician: software reads the photograph and returns a referral decision, with no clinician reading the image.
- Beside the clinician: software adds its own reading to the clinician's.
- Beyond the eye: software reads the retina for cardiovascular risk or the genetics of an inherited retinal disease.
- In refraction and surgical planning: a deep learning autorefractor, and an intraocular lens (IOL) calculator.
- With the patient: a chatbot that answers questions about an operation.
Data as a table
| Point in the examination | Tool | What was tested | Result | Randomised trial of the tool |
|---|---|---|---|---|
| Before the clinician | Autonomous diabetic retinopathy screening | ACCESS trial, 164 young people with diabetes | Eye examination completed within 6 months: 100 vs 22 percent | Yes |
| Beside the clinician | RetCAD and SELENA+, Pilbara van | 78 patients, nine communities, a remote clinician supervising, no comparator arm | Eleven times as many screened as by local primary care cameras in the same months of 2023 | No |
| Beside the clinician | EyeFM copilot | Randomised trial, 668 participants, 16 ophthalmologists | Correct diagnosis: 92.2 vs 75.4 percent | Yes |
| Beside the clinician | Hong Kong AI-OCT for macular oedema | Randomised trial, 276 referred patients | False-positive referrals: 24.1 vs 69.1 percent | Yes |
| Beside the clinician | Octane OCT referral model (HERMES) | Community optometry referrals, 396 participants | Processed images from 51.5 percent; accuracy similar to the optometrists | No |
| Beside the clinician | RetInSight Fluid and GA Monitors | On the Australian register | No trial found | No |
| Beyond the eye | Retina4IRD | Randomised trial, 300 people | Correct genetic category in the top five: 88.5 vs 67.3 percent | Yes |
| Beyond the eye | Cardiovascular risk from the retina | On the Australian register (Mediwhale, Eyetelligence) | No randomised trial found | No |
| Refraction and surgical planning | Tongren Digital Sight autorefractor | Randomised crossover trial, 282 adults | Mean difference from subjective refraction: 0.07 dioptres | Yes |
| Refraction and surgical planning | Zeiss IOL calculator | On the Australian register | No trial found | No |
| With the patient | Cataract surgery chatbot | Pilot randomised trial, 64 patients | No significant difference detected | Yes |
Before the clinician
Four systems are cleared by the United States (US) Food and Drug Administration (FDA) to make the decision without a clinician. They are LumineticsCore from Digital Diagnostics, formerly IDx-DR, EyeArt from Eyenuk, AEYE-DS from AEYE Health, and iPredict-DR from iHealthscreen, which was cleared on 2 July 2026 [12, 13].
The gain is that the examination happens. Wolf and colleagues offered young people with diabetes an AI eye examination in their own diabetes clinic. Every one of them completed it within six months. Of those given a scripted referral to an eye care provider and education, 22 percent did [4]. Issue 3 follows these systems out of their pivotal trials.
Beside the clinician
The tool shows its own reading beside the clinician's, and the clinician still signs the decision.
- EyeFM is a model pretrained on 14.5 million eye images paired with clinical text. In a single-centre, double-masked randomised trial in a high-risk Chinese screening population, 668 participants were allocated to 16 ophthalmologists, half working with the copilot. Ophthalmologists with it reached the correct diagnosis in 92.2 percent of cases against 75.4 percent, and the correct referral in 92.2 against 80.5 percent. At follow-up, 33.7 percent of patients in the copilot arm had acted on a referral suggestion, against 20.2 percent [5].
- The Hong Kong AI system adds a macular oedema reading from optical coherence tomography (OCT) to the photograph report a screening programme already produces. A multicentre randomised non-inferiority trial enrolled 276 referred patients. Referring on both reports cut false-positive referrals to 24.1 percent, against 69.1 percent on the photograph report alone. Sensitivity was 100 percent in both arms [6]. Every participant still saw the specialist, so the trial compared referral decisions and nothing downstream of them.
- Two RetInSight products on the Australian register measure rather than diagnose. One measures fluid volumes in neovascular age-related macular degeneration. The other measures the area of geographic atrophy (GA). The register says the results should not be the sole basis for a diagnosis or treatment decision [11]. I found no trial that randomised patients to care guided by either.
- In the United Kingdom HERMES study, an OCT referral model called Octane was tested on its own against expert decisions in community optometry. It was not used in care. It could process images from 204 of 396 participants, 51.5 percent. On referral decisions it was less accurate than hospital experts and about as accurate as the community optometrists [7].
Issue 2 covers two studies of what a second reading does to the clinician's own judgement, including when the AI is wrong.
Beyond the eye
- Mediwhale's software on the Australian register takes a fundus image, indicates suspected cataract, glaucoma and retinal disease, and predicts cardiovascular risk. Both Eyetelligence entries on the register also carry a cardiovascular risk output [11]. I found no randomised trial of cardiovascular risk read from the retina, so what it changes for a patient is not yet known.
- Retina4IRD reads colour fundus photographs and OCT scans and predicts which of 17 genotype categories lies behind a suspected inherited retinal disease, before genetic testing. A multicentre trial randomised 300 people to a specialist with or without it. The correct genetic category was in the specialist's top five in 88.5 percent of cases with the tool, against 67.3 percent without [8].
Refraction and surgical planning
- Tongren Digital Sight is a portable autorefractor: an infrared camera and a deep learning model trained on 362,000 images. A randomised crossover trial tested it in 282 adults in Beijing. Its spherical equivalent differed from subjective refraction by a mean of 0.07 dioptres, and corrected acuity with its prescription was non-inferior on the day. Agreement was weaker in high myopia [9].
- A Zeiss calculator on the Australian register predicts the residual refraction after surgery for a given eye biometry and IOL power [11]. No trial of it appears in the search window.
With the patient
A cataract surgery chatbot at the university hospital in Brussels answered only with clinician-validated responses. In a pilot randomised trial of 64 patients, brochures plus the chatbot made no detectable difference to knowledge, anxiety or satisfaction compared with brochures alone. Only 35 of the 64 returned the questionnaire, and 17 of the 33 given the chatbot never used it [10]. A pilot this small cannot show that the two approaches are equivalent either.
What is available in Australia
The Therapeutic Goods Administration (TGA) publishes a list of AI-enabled devices in the Australian Register of Therapeutic Goods (ARTG). Updated 10 September 2026, it carries eight ophthalmic entries [11]. Four appear above: Mediwhale's software, RetInSight's two monitors and the Zeiss calculator. The other four are these.
- TeleMedC's DR Grader screens and grades diabetic retinopathy from colour fundus photographs.
- The iCare RETCAD software reads colour fundus photographs for age-related macular degeneration, diabetic retinopathy and glaucoma, as decision support.
- Two Eyetelligence products classify age-related macular degeneration, diabetic retinopathy, glaucoma and cardiovascular risk from colour fundus photographs.
Only TeleMedC's entry describes automatic grading, and none says the software decides without a clinician. The four FDA-cleared autonomous systems are not on the list. I searched the public ARTG for each of them and its maker on 30 September 2026 and found no matching entry [14].
That is not proof of absence. The TGA's list includes a device only if AI terms appear "in the product name or intended purpose" and the sponsor reported this after March 2025 [11]. SELENA+, a Singapore retinopathy grader, was one of the two systems in the Pilbara van. It has held ARTG entry 418401 since 17 August 2023 without being on the list [15]. A name search cannot rule out an entry under a different product name or sponsor.
The rule that changes on 1 November
From 1 November 2026, new rules apply to clinical decision support software (CDSS). It can be exempt from inclusion in the ARTG only if it meets five criteria [16].
It must make recommendations to a health professional, and must not process an image or signal from another device. It must not replace the clinician's judgement or make a diagnosis or treatment decision itself. And it must display the guidelines, calculations or logic behind its recommendations, so that the clinician can "readily interpret and verify" them while using it.
The TGA revised its guidance on 16 October 2026 and says it is under review to match. The revision removed the mention of AI from the guidance's section on the exemption criteria. The guidance now says that opaque software is not exempt, and that software "could be opaque regardless of whether it incorporates technologies such as machine learning" [17]. Fixed rules do not make a product exempt by themselves. The next issue covers the rules in full.
Conflict of interest, funding and how this was written
I own OptoGuide Pty Ltd, which sells OptoGuide, a clinical decision support product for optometrists. Its clinical recommendations come from fixed rules, and no AI model runs when a clinician uses it.
No one funded this issue. I have no financial relationship with Lions Outback Vision, Ninox Vision, Google, or any maker of a system named here. I checked every figure against the paper or the register it is cited to, and the errors are mine. This series will cover the field without promoting OptoGuide, and where a finding bears on rule-based decision support I will say so.
The next issue covers the rules for Australian optometrists using AI: the guidance from the Australian Health Practitioner Regulation Agency (Ahpra) and the Code of conduct, the TGA's treatment of software as a medical device, the privacy obligations, including the automated decision-making transparency requirement that starts on 10 December 2026, and where liability sits when a tool is wrong.