Artificial intelligence in Ophthalmology: Where do we stand?

Artificial intelligence (AI) has entered everyone’s lives in multiple fields. Commendable progress has been achieved for the integration of AI in the ophthalmic field, and a few of its challenges were highlighted in the previous editorial.^[1] While the use of AI is primarily restricted to diagnosis, suggestion of treatment and predicting the course of the diseases over time, this editorial is aimed to give an overview of certain diseases wherein AI will play an important role in the times to come, in the field of Ophthalmology.^[2-4]

Certain terminologies are wrongly used interchangeably. They are AI, machine learning (ML) and deep learning (DL). The AI was conceptualised by Savastano MC et al. as an intelligent machine with consistent human-like problem-solving ability.^[2] It works on specialised computers based on data and certain algorithms. It worked repeatedly without deviation and is known as ML. The three main approaches are (a) unsupervised learning, (b) supervised learning and (c) reinforcement learning.

DL, on the other hand, is the replication of the human brain, which constantly learns from exposure to data and improves the accuracy of the output. Generative adversarial networks, Generative AI, Recurrent neural networks, Convolutional neural networks, Large language models and Deep neural networks all help in the development of DL. Hence, DL uses several processing layers to draw advanced features from the input data and its algorithms, which are known as ‘black boxes’. These features are possible due to the ever-increasing widespread availability of ‘big data’ and ultra-rapid processing through supercomputers, in miniscule sizes, at a very reasonable cost.

In the speciality of ophthalmology, AI has shown tremendous potential due to the availability and integration of various imaging modalities such as fundus photography, optical coherence tomography (OCT) and visual field recording. AI has also proved to be helpful in situations where enormous data need to be incorporated in getting accurate predictions, such as intraocular lens power calculation, preoperatively. Although a detailed discussion is out of scope in this editorial, I will mention some of the conditions wherein AI holds a good promise in the coming years. Subclinical keratoconus identification can save many eyes from developing iatrogenic keratectasia in future, after refractive surgery. Dry eye disease holds a bright future with regard to its management. The prevention of blindness due to retinopathy of prematurity is possible if diagnosed early and treated properly. AI can help in the swift and accurate detection of this condition. Time plays an important role in the detection of diabetic retinopathy, and it is not difficult to see that the use of AI has proved to be very helpful in taking fundus pictures of diabetics by technicians using a mobile-based camera. The technicians simply send these pictures to the reader centre, wherein the patients who need further treatment are triaged and sent to higher centres. Similarly, the use of AI in age-related macular degeneration also has a great potential where classification and prediction of progression in patients who are at a higher risk of blindness can be identified and monitored.

Glaucoma is known to be erroneously managed by majority, and the ‘diagnosis’ as well ‘choice of treatment’ remains to be a great bugbear. Ongoing research indicates that AI has a huge potential in solving this riddle by integrating the fundus picture, OCT measurements and perimetry. Once incorporated in routine clinical practice, AI will help match the clinical efficiency of common practitioners with the best in the field.

The ‘eye’ is the window of the body. We are aware that certain diseases can be predicted based on the changes in the retina. One can comfortably say that AI will widen this field beyond limits due to its ability to detect subtle changes in the vasculature and retinal layers, which would otherwise not have been possible to detect through the human eye. Cardiovascular disease, certain neurological conditions and chances of developing Alzheimer’s disease are some conditions which can be predicted by ocular examination with the help of AI.

It will be fair to say that though AI holds a promising place in the medical field, the role of a physician will never be outdated in total management. This role will necessarily have to be a balancing act by the physicians, is something which will evolve with the advancement of AI in the years to come.