A Web Base Application for Keratoconus Detection

Abstract

Keratoconus, a common eye disorder affecting approximately one in 2,000 individuals worldwide, leads to a significant decline in visual acuity. Early diagnosis is pivotal to halting its progression and preserve vision. The study introduces a machine learning algorithm designed for the early detection of keratoconus, utilizing a comprehensive dataset comprising corneal parameters obtained from diverse clinical sources. The Federal University of Sao Paulo’s Institutional Review Board has provided ethical approval in ensuring adherence to ethical principles and the Statement of Helsinki. The dataset was meticulously de-identified to safeguard participant privacy. Employing a range of machine learning models, diagnostic accuracies ranging from 56.57% to an impressive 93.65 % have been identified. This pioneering model supports doctors in evaluating corneal health and keratoconus detection, simultaneously, at the early, preclinical phases of the illness, where subjective evaluations often fall short. Additionally, there is potential for seamless integration of this algorithm into existing methods to capture images of the cornea or software specifically designed for this purpose, significantly enhancing keratoconus detection capabilities.