Varsha Singh, B. Optom.
M. Optom Student, The Sankara Nethralaya Academy, Chennai
Artificial Intelligence (AI) is transforming the future of eye care by providing viable solutions to the current barriers in eye care, such as inaccurate diagnoses and a lack of manpower. Day by day, it is gaining popularity as it enhances eye care practice by reducing the time required for disease screening, detection, diagnosis, and management, and by improving diagnostic accuracy. The option of AI for eye care is available from basic refraction to advanced retinal diagnosis. (1)
Recent Advancements in AI Tools for Eye Care
Automated Retinoscopy
The AI-powered, portable retinoscopy system uses a smartphone attached to the standard retinoscope to record the video and estimate the net refractive error by using a modified retinoscopy algorithm. The reported sensitivity and specificity values are 91% and 74% respectively, with the Mean Absolute Refractive Error (MAE) compared to subjective refraction being 0.75 ± 0.67 D. (2)
| Feature | Description |
|---|---|
| Technology | AI-powered portable retinoscopy system using a smartphone attached to a retinoscope. |
| Principle | Records video during retinoscopy → applies modified retinoscopy algorithm → estimates net refractive error. |
| Performance | Sensitivity: 91%, Specificity: 74%, Mean Absolute Error (MAE): 0.75 ± 0.67 D compared to subjective refraction. |
| Applications | Screening and diagnosis of refractive errors in community eye care and low-resource settings. |
| Advantages | Portable, inexpensive compared to autorefractors, reduces manpower dependency. |
| Limitations | Less precise than subjective refraction, may require stable fixation and good video quality. |
Table 1: Features of Automated Retinoscopy
AI-Driven Personalisation in Myopia Management
Fundus2Globe is an AI tool that constructs a 3D model of the human eye by taking 2D inputs, which is useful in predicting disease progression and in deciding personalised treatment strategies. (3)
CeViT (Copula-Enhanced Vision Transformer) is a deep learning architecture designed for image-based myopia screening using ultra-widefield (UWF) fundus images. It also classifies high myopia and predicts axial length elongation. (4)
| Feature | Description |
|---|---|
| Technology | Deep learning model (Vision Transformer architecture enhanced with copula-based dependency modeling). |
| Principle | Uses ultra-widefield (UWF) fundus images to classify myopia severity and predict axial length elongation. |
| Performance | High accuracy in detecting high myopia. |
| Applications | Population-based screening programs, clinical myopia monitoring, prediction of progression. |
| Advantages | Handles large-field imaging better than conventional CNNs, enabling early intervention. |
| Limitations | Requires UWF devices (expensive), not yet widely validated in all populations. |
Table 2: Features of CeViT
Anterior Segment Evaluation
CorneAI is an AI-based deep learning model that is developed to improve the diagnostic accuracy of the anterior segment, with an accuracy of 86%. It is also reported that it has improved the diagnostic accuracy by 9.6%, and when compared to the slit-lamp, the diagnostic accuracy improved by 9.4%. (5)
KerNet is the model to detect the asymmetric keratoconic eye by utilising the data from the maps provided by Pentacam HR, with the reported accuracy of 94.12%. (6)
| Feature | Description |
|---|---|
| Technology | AI-based model for keratoconus detection. |
| Principle | Utilises Pentacam HR maps (topography & tomography) to differentiate asymmetric keratoconic eyes. |
| Performance | Accuracy: 94.12%. |
| Applications | Early detection of keratoconus, screening candidates for refractive surgery. |
| Advantages | More sensitive in detecting subtle asymmetry compared to human grading. |
| Limitations | Pentacam required (costly), limited to keratoconus and not other anterior segment pathologies. |
Table 3: Features of KerNet
Retinal Diagnostics and Screening
IDx‑DR and AEYE are AI software programs that are FDA-approved. They are designed particularly for the detection of diabetic retinopathy (including macular oedema). They do not require an eye care professional to interpret the result; they will automatically give you the diagnosis. (7,8)
| Feature | IDx-DR | AEYE |
|---|---|---|
| Approval | First FDA-approved autonomous AI for DR (2018) | FDA-approved AI for DR |
| Input | Fundus photographs | Fundus photographs |
| Output | Detects referable DR ± macular edema | Detects referable DR ± macular edema |
| Mode | Standalone, offline certified system | Cloud-based, integrates with workflows |
| Performance | Sensitivity & specificity >85–90% | Sensitivity & specificity >85–90% |
| Advantages | Proven regulatory credibility, no clinician needed | Faster reporting, scalable, remote use |
| Limitations | DR only, high-quality images needed, limited availability | DR only, internet dependent, limited global access |
Table 4: Features of IDx-DR and AEYE
EyeFound and GlobeReady are the multimodal foundation AI models in the field of eyecare to diagnose multiple retinal diseases by utilising Colour Fundus Photographs (CFPs) and OCT images. The reported accuracy ranges from 94.9%-99.4% for CFPs and 88.2%–96.2% for OCT images, respectively, for the GlobeReady model. (9,10)
| Feature | GlobeReady | EyeFound |
|---|---|---|
| Type | Multimodal foundation AI model | Multimodal foundation AI model |
| Input | Fundus photographs (CFPs) + OCT images | Fundus photographs (CFPs) |
| Output | Detects multiple retinal diseases using combined CFP + OCT data | Detects multiple retinal diseases from CFPs |
| Performance | Accuracy: 94.9–99.4% (CFPs), 88.2–96.2% (OCTs) | Accuracy: 94.9–99.4% (CFPs) |
| Applications | Comprehensive retinal screening (DR, AMD, glaucoma, etc.) | Broad retinal disease screening (mainly CFP-based) |
| Advantages | Integrates structural + functional imaging, higher reliability | High accuracy from single-modality input, scalable |
| Limitations | Requires OCT (expensive, less accessible), high computational demand | Limited to CFPs only, may miss OCT-based features |
Table 5: Features of GlobeReady and EyeFound
Conclusion
There are various AI tools and models available, and new modalities are emerging, providing several advantages in eye care, be it early detection, time-saving screening, or accurate diagnoses. However, most of these are not easily accessible and remain confined to research laboratories. With the innovation, attention should also be directed toward the accessibility aspect of these tools and models by considering all these points. (1,11)
References
- Krishnan, A., Dutta, A., Srivastava, A., Konda, N., & Prakasam, R. K. (2025). Artificial Intelligence in Optometry: Current and Future Perspectives. Clinical optometry, 17, 83–114.
- Aggarwal, A., Gairola, S., Upadhyay, U., Vasishta, A. P., Rao, D., Goyal, A., … Jain, M. (2022). Towards automating retinoscopy for refractive error diagnosis.
- Shi, D., Liu, B., Tian, Z., Wu, Y., Yang, J., Chen, R., … He, M. (2025). Fundus2Globe: Generative AI-driven 3D digital twins for personalized myopia management.
- Zhong, C., Li, Y., Xu, J., Fu, X., Liu, Y., Huang, Q., … Liu, C. C. (2025). CeViT: Copula-Enhanced Vision Transformer in multi-task learning and bi-group image covariates with an application to myopia screening.
- Maehara, H., Ueno, Y., Yamaguchi, T., Kitaguchi, Y., Miyazaki, D., Nejima, R., … Oshika, T. (2025). Artificial intelligence support improves diagnosis accuracy in anterior segment eye diseases. Scientific Reports, 15(1), 5117.
- Xu, Z., Feng, R., Jin, X., Hu, H., Ni, S., Xu, W., … Yao, K. (2022). Evaluation of artificial intelligence models for the detection of asymmetric keratoconus eyes using Scheimpflug tomography. Clinical & Experimental Ophthalmology, 50(7), 714–723.
- Cornwell, K. (n.d.). Artificial intelligence and emerging technologies in optometry and ophthalmology. EyesOnEyecare.
- Healio Staff. (2024). Top AI stories of 2024: New opportunities for screening innovation. Healio: Optometry.
- Shi, D., Zhang, W., Chen, X., Liu, Y., Yang, J., Huang, S., … He, M. (2024). EyeFound: A multimodal generalist foundation model for ophthalmic imaging.
- Wang, M., Lin, T., Hou, Q., Lin, A., Wang, J., Peng, Q., … Cheng, C.-Y. (2025). A clinician-friendly platform for ophthalmic image analysis without technical barriers.
- Blandford, A., Abdi, S., Aristidou, A., Carmichael, J., Cappellaro, G., Hussain, R., & Balaskas, K. (2022). Protocol for a qualitative study to explore acceptability, barriers and facilitators of the implementation of new teleophthalmology technologies between community optometry practices and hospital eye services. BMJ open, 12(7), e060810.