Artificial Intelligence in Cornea, Refractive, and Cataract Surgery

Aazim A. Siddiqui; John G. Ladas; Jimmy K. Lee


Curr Opin Ophthalmol. 2020;31(4):253-260. 

In This Article

Artificial Intelligence and Keratoconus

Artificial intelligence can play a significant role in corneal disorders, such as keratoconus and other ectatic diseases. The process of screening, diagnosing, and managing diseases, such as keratoconus, can be significantly improved upon with the help of artificial intelligence.[6]

Keratoconus is characterized by progressive, often bilateral, thinning, protrusion, and scarring of the cornea. It can initially be managed conservatively with refractive correction and rigid gas permeable contact lenses. But more advanced forms of the disease require more invasive modalities of treatment such as intracorneal ring implantation and corneal collagen cross-linking. The most invasive form of treatment is corneal transplantation, which is reserved for severe cases of keratoconus.[6]

A large part of the diagnosis and management of keratoconus revolves around prevention and halting disease progression. In such cases, early screening and detection is imperative. Further, early detection can provide necessary preoperative information as an indication of a patient's risk of postlaser in-situ keratomileusis (LASIK) complications. For most ophthalmologists, detecting and diagnosing clinical ectasias that are not clinically apparent remains a challenge. This can be because of the subjective, often time-consuming and unreliable interpretation of topographic and tomographic maps of the cornea. This leads to subjective decision-making, which may not provide accuracy in appropriately identifying early ectatic changes.[6]

Modern corneal topographic and tomographic mapping yields a vast quantity of variables and datapoints. This data can serve as the perfect landscape to train an artificial intelligence algorithm and potentially yield an alternative solution to identifying subjects of interest. Recent trials have demonstrated the use of Orbscan IIz (Bausch & Lomb, Rochester, New York, USA) and the data obtained from it to highlight corneal astigmatic aberrations. This data was classified in various machine learning algorithms and radial basis neural network functions.[6] This was successful in detecting the various corneal abnormalities, which could potentially be clinically beneficial. These algorithms were found to be similar to nonartificial intelligence indices, such as KISA% index, Klyce/Maeda Keratoconus Index, and Cone Location, and Magnitude Index.[6]

Following these earlier research studies, subsequent studies found the Scheimpflug tomography-based devices to be superior to the Orbscan IIz. This was because of improved anterior and posterior surface topography measurements.[6] Various studies have performed comparisons between normal and keratoconic topographic and tomographic maps, which have allowed for further refinement. On the basis of the results of these research studies, one may conclude that there is a role for artificial intelligence to play in identifying various features of corneal ectasias and thereby playing a major role in the diagnostic process.