Dr. Abhishek Mandal, Ph.D.

Senior Business Adviser, Vision Science Academy, London, U.K.

Vision Science Academy Exclusive


What are Convolutional Neural Networks?

Convolutional neural networks (CNN) are a modern artificial intelligence (AI) tool, which has been widely utilized to conduct a visual image analysis. CNNs have been implemented within diverse fields of medicine, where deep learning can automatically decipher advanced clinical features of a multitude of pathological conditions. In this way, deep learning tools such as CNNs allow an early diagnosis and staging of a medical disorder.

A CNN comprises different layers i.e., convolutions, pooling, and fully connected layers. The output of each layer can be related to the input of another, and the final output gives us a valuable clue regarding the disease classification.

Application of CNN in Vision Science

Computer guided technology has revolutionized different aspects of eye care. Detailed retinal images can be created by using AI softwares, and interpretation of this data allows a highly sophisticated identification of the underlying disease (Tong, Lu, Yu, & Shen, 2020). Following are some examples of CNN application in eye care:

CNN in the Diagnosis of Glaucoma

Glaucoma is one of the most notorious causes of blindness in the developing world. Complex AI networks have been designed to achieve an autonomous and reliable diagnosis of glaucoma. CNNs capture hundreds of fundus images and then deploy their algorithm which analyses retinal features such as optic disc size, cup-disc ratio, retinal nerve fibre layer thickness, and neuroretinal rim thickness, and allows a rapid identification of glaucomatous eye (Wang et al., 2019). Such CNN models have a specificity and sensitivity exceeding 90% which makes them ideal for use in clinical settings. Nonetheless, factors such as population variability can potentially lead to lower levels of diagnostic accuracy (Gheisari et al., 2021).

CNN in the Diagnosis of Diabetic Retinopathy

Fundus imaging has also allowed CNN to monitor the progression of diabetic retinopathy (DR) (Mohamed Shaban et al., 2020). DR is a major microvascular complication of diabetes which can gradually lead to blindness. Although the initial stages of the disease are relatively asymptomatic, it is nonetheless, imperative to diagnose DR at the earliest stage possible. Various CNN models have been developed which analyse characteristic features of DR such as hard exudates and morphological changes in the retinal vessels including micro-aneurysms. High levels of specificity and sensitivity have been obtained in this regard both of which exceed 90% (Nayak, Bhat, Acharya, Lim, & Kagathi, 2008). Furthermore, DR has been successfully classified into different stages by using CNN models with a high accuracy of nearly 80% (M. Shaban et al., 2018).

CNN in the Diagnosis of Age-Related Macular Degeneration

Age-related macular degeneration (ARMD) is a major cause of blindness among the elderly population. Only an early diagnosis can help improve the overall disease prognosis. Researchers have developed an advanced CNN system which can help analyse retinal changes associated with ARMD during early phases with an accuracy as high as 90%. Such cost-effective modalities can certainly help improve the living standards of patients who are visually impaired due to ARMD (Tan et al., 2018).

Future Directions

More resources need to be invested for the implementation of advanced CNN systems in eye care since this form of AI has the potential to revolutionize the field of diagnostic medicine.



Gheisari, S., Shariflou, S., Phu, J., Kennedy, P. J., Agar, A., Kalloniatis, M., & Golzan, S. M. (2021). A combined convolutional         and recurrent neural network for enhanced glaucoma detection. Sci Rep, 11(1), 1945. doi:10.1038/s41598-021-81554-4

Nayak, J., Bhat, P. S., Acharya, R., Lim, C. M., & Kagathi, M. (2008). Automated identification of diabetic retinopathy stages         using digital fundus images. J Med Syst, 32(2), 107-115. doi:10.1007/s10916-007-9113-9

Shaban, M., Ogur, Z., Mahmoud, A., Switala, A., Shalaby, A., Abu Khalifeh, H., . . . El-Baz, A. S. (2020). A convolutional neural         network for the screening and staging of diabetic retinopathy. PLoS One, 15(6), e0233514.         doi:10.1371/journal.pone.0233514

Shaban, M., Ogur, Z., Shalaby, A., Mahmoud, A., Ghazal, M., Sandhu, H., . . . El-Baz, A. (2018, 6-8 Dec. 2018). Automated         Staging of Diabetic Retinopathy Using a 2D Convolutional Neural Network. Paper presented at the 2018 IEEE         International Symposium on Signal Processing and Information Technology (ISSPIT).

Tan, J. H., Bhandary, S. V., Sivaprasad, S., Hagiwara, Y., Bagchi, A., Raghavendra, U., . . . Acharya, U. R. (2018). Age-related         Macular Degeneration detection using deep convolutional neural network. Future Generation Computer Systems, 87,         127-135. doi:https://doi.org/10.1016/j.future.2018.05.001

Tong, Y., Lu, W., Yu, Y., & Shen, Y. (2020). Application of machine learning in ophthalmic imaging modalities. Eye and         Vision, 7(1), 22. doi:10.1186/s40662-020-00183-6

Wang, P., Shen, J., Chang, R., Moloney, M., Torres, M., Burkemper, B., . . . Richter, G. M. (2019). Machine Learning Models for         Diagnosing Glaucoma from Retinal Nerve Fiber Layer Thickness Maps. Ophthalmol Glaucoma, 2(6), 422-428.         doi:10.1016/j.ogla.2019.08.004