An ophthalmologist made AI?
What can an ophthalmologist do with AI?

Over eight million smart speakers were sold by the end of 2019.
Yes, we too use AI-based smart speakers. We call the speaker as soon as we wake up in the morning. However, do we have to use the AI to do this? We can do this easily by a few movements of the fingers on a smartphone. These are all different forms of AI too. Regardless of whether you ask or not, the AI does the task for you. It is amazing and convenient that they do it for you, even though you did not know you needed it and did not ask for it. Then, can it respond to things you did not ask? There are things we must always ask and respond ourselves. For example, the patients’ vision after a SMILE surgery, lens size befitting the patient for the ICL implantation, and the optical power of the intraocular lens during cataract surgery. Not only doctors, but also counselors must ask themselves endlessly by looking at the accumulating checkup results while the patients are in the process of being checked up.

- Is refractive surgery possible for this person?
- Then what surgical option is appropriate?
- What will the vision be after surgery?

Is it possible for AI to respond to such questions? Then, we can consider the patients’ perspective and interact with them more deeply and warmly.

Why doctors talk about figures
Doctors typically talk about figures (in other words, numerical digits), not because they like it, but because figures are an important part of the treatment and visual correction. Here is an example. “This patient has a thick cornea, and the right and left myopia is -6.50 diopter and the with-the-rule astigmatism* is 2.00 diopter; therefore, if the figures are entered in the surgery nomogram as such and a femtosecond laser is used to a particular degree, the myopia should disappear after surgery, following which the predicted visual acuity would be 1.2 in the decimal scale. Since the xerophthalmia is low and pupil size is relatively small, the associated inconvenience after surgery is not expected to be big.” *The myopia that leads to blurry vision, due to the lack of correct focus, can be classified broadly into with-the-rule astigmatism and against-the-rule astigmatism depending on the location of the focal line with the strongest refraction.

When consulting the patient to receive visual correction through SMILE surgery, it is crucial that the doctor promptly checks the preoperative measurement results with the eye, examines the patient’s eyes with slit lamp microscope, and calculates the surgical values. To ensure a precise and safe surgery, the calculations need to be repeatedly verified. Since the doctor must concentrate on precisely calculating each figure in the presence of the patient; furthermore, the doctor has no reserve power to listen to all the patients’ stories. However, if the AI would be tasked with the calculations, the doctor would be able to focus on the patient’s eyes, rather than at the monitor. It is almost as if the eyes of the patient were looking at my eyes to meet a bright world. Since a precise calculation that was performed by a machine would be better than those performed by humans, the several calculations required for visual correction surgery can also be better performed by AI. This is considering if AI is taught with our knowledge and experience. Recently, rather than AI, which primarily replaces the human intelligence, augmented intelligence was conceptualized. This aims to supplement human intelligence rather than replace it. Therefore, the latter AI will be of substantial aid to solving our problems.

Why ophthalmologists see the image?
An important ophthalmological characteristic makes us consider AI. Unlike other treatment subjects that show the diagnosis results with a few figures, most ophthalmological evaluation involves complex figures and images. Deciphering the evaluation results is similar to diagnostic radiology. However, unlike black and white x-ray, MRI and CT, the ophthalmological evaluation also includes several 3D and colored images. Therefore, ophthalmology is an area where AI is most actively being researched. Analyzing just one eye constitutes data of approximately 70 images; however, in most situations both eyes require investigation and analysis. Therefore, we believe this sufficiently explains the amount of image data that is produced for an ophthalmologist to review before treatment and for planning the surgery.

Considering these characteristics of ophthalmology, many prominent global firms such as Google are focusing on developing ophthalmological AI to judge the systematic health state, as well as the retina, through fundus photography (1). Generally, the Pentacam is the representative checkup of an ophthalmologist. It can precisely check the state of the cornea before surgery and is a checkup tool that can diagnose conical cornea, which is a condition that should not receive visual correction surgery. This checkup is deciphered through 3D and high-quality color images and figures of the frontal region of the eyes such as the cornea and the anterior chamber.

Slit lamp and corneal topography images of a normal cornea

This checkup result, which looks like a sandglass, indicates a normal corneal shape, which is appropriate for surgery.

Slit lamp and corneal topography images of an abnormal cornea

In contrast, in a case as that shown above with an irregular or horse’s hoof shaped cornea, general laser visual correction surgery is preferably avoided, but customized laser surgery must be considered.
An experienced visual correction ophthalmologist can intuitively decide by seeing such color images, shapes, and figures, whether it is a normal eye or an eye that can undergo visual correction surgery. Greater amount of experience is associated with a deeper judgment and understanding of treatment modality and checkup outcome. Therefore, a higher level of experience has been correlated with a more precise treatment and surgery planning.

However, judging and interpreting complex figures and images from preoperative examinations is precisely what AI can also perform with a considerable degree of accuracy. If a large amount of data is used to train the models, the machine learning based AI can provide an output in one attempt and categorize into a normal eye where general laser ablation surgery is possible, or an eye that requires a customized surgical approach or an eye that should not receive surgery.

A schematic illustrating the logical process to identify the candidates for corneal refractive surgery.

What ophthalmologists do with AI?

If the AI that has learnt the demographic information of the patient receiving surgery and clinical decision-making database of the doctor were to acquire the checkup results of the new patient, it can help the doctor by assisting the system. Specifically, what one can expect from the AI is the precision of judgment that is expected of highly skilled doctors and counselors. This indicates that an additional trustworthy basis of judgment has been made for doctors and patients.
Then, is this good for the doctors? Or is it good for the patients? It is good for patients because they can be recommended, the precise surgical methods that are safe and personalized. It is also good for doctors that are lacking in knowledge and experience as it eliminates the proficiency gap. Therefore, does it mean that it is not particularly needed for experienced doctors? Although it may not be necessary, it could act as an added advantage. Considering the fact that doctors too, are humans, and they can make human errors, due to either their emotions or fatigue. Here, AI can support doctors as a very objective revalidation system.

That is why we decided to design an AI. We have designed it based on a large quanta of data of 420,000 eyes and will continue to elaborate on and expand it. We also published a research article using our AI system . The B&VIIT’s research article demonstrating that machine learning can accurately identify candidate patients for refractive surgery. Globally, several researchers have been interested in the B&VIIT’s article.