Table 1.

Novel applications of artificial intelligence (AI) in keratoconus management

Potential application
1- Screening: Machine learning techniques could differentiate normal and keratoconic eyes, as well normal and form-fruste keratoconus [79]. An AI-based model with simultaneous use of Scheimpflug and Placido corneal imaging data revealed a high diagnostic accuracy for early detection of keratoconus and discriminated keratoconic eyes from normal corneas [80]. AI-based classifiers were helpful in detecting early keratoconus, but cannot replace clinical experts’ opinions, notably in decision-making before refractive surgery. However, experts’ opinions are not error-free [81]. An AI-based algorithm discriminated accurately between normal, suspect irregular, and keratoconic corneas, similar to a corneal expert. The authors recommended applying machine learning to corneal tomography to ficilitate screening of keratoconus in large populations and refractive surgery candidates [82].
2- Diagnosis A machine learning-based algorithm could improve diagnosis of subclinical keratoconus or early keratoconus in routine ophthalmic practice. However, no consensus has been reached concerning the corneal parameters that should be included for assessment, or the optimal design [83]. Combining segmental tomography with Zernike polynomials and AI achieved an excellent classification of healthy and keratoconic eyes. The AI model efficiently classified the eyes with very asymmetric ectasia as subclinical and forme-fruste keratoconus [84]. The keratoconus classification performance of a random forest classifier combined with sequential forward selection method achieved a high accuracy with a reduced execution time [85]. An AI-based diagnostic model using purely biomechanical parameters without corneal topographic examination, demonstrated a rapid and accurate diagnostic performance for keratoconus [86].
3- Treatment AI-based diagnostic software that included a model for the automated determination of keratoconus stage achieved an accuracy from 0.95 to 1.00 relative to the adapted Amsler–Krumeich algorithm. The software contained a standardized algorithm for determining surgical intervention indications, drived from data available in the literature and recommendations from the expert community [87]. Compared with the manufacturer’s nomograms an artificial neural network revealed better performance regarding better corrected vision and lowering of the coma-like aberrations in guiding intracorneal ring segments implantation in eyes with keratoconus [88]. Corneal crosslinking in a 28-year-old patient with bilateral keratoconus combined with photorefractive surface ablation customized by a novel AI-based platform for calculating lower- and higher-order aberrations based on wavefront data, Scheimpflug tomography, and interferometry-based axial length measurements from a single diagnostic device achieved an accurate normalization of distorted eye optics [89].
4- Patient Follow-up The orthopedic field developed an AI-assisted follow-up system for post-operative monitoring. Its effectiveness was not inferior to that of manual follow-up and led to saving human resource costs. [90]. Similar verified AI-assisted follow-up systems, modified for compatibility with the field of ophthalmology, could be developed to monitor patients with keratoconus post-operatively.
5- Patient Rehabilitation Goodman and Zhu considered successful examples in the ophthalmology literature and proposed potential use of AI-based algorithms for personalized surgical planning to improve postoperative outcomes and visual rehabilitation in patients with keratoconus [91]. Studies on AI-based algorithms for delivery of home-based virtual rehabilitation programs to adult patients revealed that incorporating AI with home-based virtual rehabilitation improved rehabilitation outcomes. The authors recommended further assessment of the effectiveness of various forms of AI-driven home-based virtual rehabilitation with consideration of its unique challenges and applying standardized metrics [92]. Similar verified AI-driven home-based virtual rehabilitation system, modified for compatibility with the field of ophthalmology, could be developed for adult patients with keratoconus.
6- Patient Education AI-based systems for online patient education in other fields have revealed promising results [93, 94]. Similar verified AI-based dialogue platforms, modified for compatibility with the field of ophthalmology could be developed for adult patients with keratoconus.