Janakiraman P, Bachelor in Optometry

Research Optometrist, Medical Research Foundation, Sankara Nethralaya, Chennai, India

Artificial intelligence (AI), the new era has evolved and is ruling technology in many aspects which in turn plays a vital role in the field of medicine including Ophthalmology. AI refers to a machine that imitates the human’s thought process and behavioural pattern. ^(1). AI is a branch of computer science that involves creating machines to replicate human intelligence. ^(1). The application of this forthcoming technology makes the field of ophthalmology digitalized, which prospects the accessibility, availability, and productivity of existing overall efficiency of eye care services, and it avidly focused on diseases such as Diabetic retinopathy (DR), Glaucoma, Age-related macular degeneration, retinal vein occlusion, Cataract and Retinopathy of prematurity. ^(1,2,3).

Aim of AI

• Detection of DR and glaucoma with fundus images, OCT images, visual fields, and other clinical data.
• Detection of the condition and referral before it worsens.
• It has also been applied to study various risk factors of the disease condition and the quality of life. ⁽⁴⁾

Classification of AI algorithms

Figure 1: Introduction of AI algorithms [Photo courtesy: Lu, W., Tong, Y., Yu, Y., Xing, Y., Chen, C., & Shen, Y. (2018). Applications of artificial intelligence in ophthalmology: general overview. Journal of ophthalmology, 2018.]

Machine learning

Machine learning (ML) is defined as a set of methods that automatically detect patterns in data and then incorporate this information to predict future data under uncertain conditions. Machine learning is further divided into Supervised and Unsupervised, which implies the labelling of the data provided for classification into disease and non-disease categories. ⁽⁵⁾

Deep learning

Deep learning is a subset of machine learning that uses an artificial neural network (ANN) structure which was inspired by the biological neural network (Visual cortex). ^(3,5)

Figure 2: Convolutional neural network framework. (a) One or more filter functions reduce portions of the image (green box) into mathematical representations, creating a feature map (grey circle). (b) Pooling functions combine similar statistics from extracted features. (c) Fully connected layers are used in the classification process, in which each node is connected with every other node in the preceding layer. Multiple repetitions in multiple layers yield a final output. [Photo courtesy: Zheng, C., Johnson, T. V., Garg, A., & Boland, M. V. (2019). Artificial intelligence in glaucoma. Current opinion in ophthalmology, 30(2), 97-103.]

Metrics that assess the AI algorithms

• Sensitivity
• Specificity
• Positive predictive value
• Negative predictive value
• Area under the operating curve (AUC) ⁽⁵⁾

Figure 3: Potential applications of a DL solution for DR screening. Illustration of the application of a DL solution for DR screening using imaging, comparing existing clinical practice with a fully automated AI model (replacement) and a semiautomated AI model (triage). AI indicates artificial intelligence, DL, deep learning: DR, diabetic retinopathy. [Photo courtesy: Gunasekaran, D. V., & Wong, T. Y. (2020). Artificial intelligence in ophthalmology in 2020: a technology on the Cusp for translation and implementation.]

Figure 4: Graphical representation of the structure of an artificial neural network (ANN). Input in this case is a fundus photograph with each feature for interpretation being a stimulus in the input layer. Stimuli is passed through the hidden layers which have learned features and weightings. This results in the output being recognized as DR. DR – Diabetic retinopathy. [Photo courtesy: Hogarty, D. T., Mackey, D. A., & Hewitt, A. W. (2019). Current state and future prospects of artificial intelligence in ophthalmology: a review. Clinical & experimental ophthalmology, 47(1), 128-139.]

Figure 5: Proposed system for the detection of glaucoma. [Photo courtesy: Nayak, J., Acharya, R., Bhat, P. S., Shetty, N., & Lim, T. C. (2009). Automated diagnosis of glaucoma using digital fundus images. Journal of medical systems, 33(5), 337-346.]

Needs to be addressed

• Scarcity of ophthalmologists or trained persons for grading the retinal images.
• It is difficult for people from rural areas to undergo eye examinations.
• Follow-up is required for years, which might hinder the patient’s interest in participation. ^(6).

To overcome this, automated imaging systems and software which can efficiently analyse the captured retinal images and are easily available to reach the patient. ⁽⁶⁾

Requirements of AI for real-world transformation

• Good and ample number of clinical datasets.
• Eminent business and product management team, funding, and research team to conduct further research.
• A most important aspect is to prop up by strong and capable ecosystem with all medico-legal terms, cyber-security, telecommunication support, supercomputing powers and networks, health economic analysis, and to educate the upcoming generation about the basic concepts of AI and application to make the world more technologized.⁽⁷⁾

Limitations

• Accuracy will be decreased if AI is applied to different ages, races, and ethnicity. ⁽¹⁾
• The requirement for glaucoma analysis differs from DR, as it requires OCT image, fundus image, Intraocular pressure, and visual field report.^(2,5)
• Possibility of false-positive and false-negative results if the sensitivity and specificity is less than 90%. ⁽⁵⁾
• DL algorithm is validated only for identifying a single eye disease at a time. ⁽³⁾
• To avoid misclassification
• Standardized practices should be employed
• Eschewing of change in pixels. ^(3,8)

Conclusion

AI is expected to make a great impact in healthcare by using the clinical datasets to narrow down the diagnosis and to set a clinical pattern to assist and aid the clinician inevitably. Future research is required to know the viability of algorithms applied in the clinical setting to rule out the development and its outcomes in comparison with ophthalmologic practice. ^(1,7,9)

References

1. Campbell, C. G., Ting, D. S., Keane, P. A., & Foster, P. J. (2020). The potential application of artificial intelligence for diagnosis and management of glaucoma in adults. British Medical Bulletin, 134(1), 21-33.
2. Padhy, S. K., Takkar, B., Chawla, R., & Kumar, A. (2019). Artificial intelligence in diabetic retinopathy: A natural step to the future. Indian journal of ophthalmology, 67(7), 1004.
3. Gunasekeran, D. V., & Wong, T. Y. (2020). Artificial intelligence in ophthalmology in 2020: a technology on the cusp for translation and implementation.
4. Zheng, C., Johnson, T. V., Garg, A., & Boland, M. V. (2019). Artificial intelligence in glaucoma. Current opinion in ophthalmology, 30(2), 97-103.
5. Hogarty, D. T., Mackey, D. A., & Hewitt, A. W. (2019). Current state and future prospects of artificial intelligence in ophthalmology: a review. Clinical & experimental ophthalmology, 47(1), 128-139.
6. Devalla, S. K., Liang, Z., Pham, T. H., Boote, C., Strouthidis, N. G., Thiery, A. H., & Girard, M. J. (2020). Glaucoma management in the era of artificial intelligence. British Journal of Ophthalmology, 104(3), 301-311
7. Lu, W., Tong, Y., Yu, Y., Xing, Y., Chen, C., & Shen, Y. (2018). Applications of artificial intelligence in ophthalmology: general overview. Journal of ophthalmology, 2018
8. Mayro, E. L., Wang, M., Elze, T., & Pasquale, L. R. (2020). The impact of artificial intelligence in the diagnosis and management of glaucoma. Eye, 34(1), 1-11.
9. Gulshan, V., Peng, L., Coram, M., Stumpe, M. C., Wu, D., Narayanaswamy, A., … & Webster, D. R. (2016). Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. Jama, 316(22), 2402-2410.