ESCRS - FPT02.10 - Using Artificial Intelligence (Ai) In Predicting Keratoconus (Kc) Demographics – A Novel Insight

Using Artificial Intelligence (Ai) In Predicting Keratoconus (Kc) Demographics – A Novel Insight

Published 2022 - 40th Congress of the ESCRS

Reference: FPT02.10 | Type: Free paper | DOI: 10.82333/rfn4-vt19

Authors: Dishitha Rathod* 1 , Rohit Shetty 1 , Gairik Kundu 1 , Pooja Khamar 1

1Cataract and Refractive,Narayana Nethralaya,Bangalore,India

Purpose

To identify and analyse the high-risk factors and demographic factors influencing the progression of keratoconus using an AI model. 

Setting

Narayana Nethralaya, Bangalore, India

Methods

500 eyes of 500 KC patients with ≥ 2 visits were included. Changes in anterior corneal curvature (Kmax) between 2 visits ≥ 6 months apart were studied. Random Forest (RF) classifier model was used to classify these patients. Demographic data & risk factor assessment was done with a questionnaire that included presence of eye-rubbing, IgE, vitamin D & B12 levels, duration of indoor activity, usage of lubricants & immunomodulator topical medications, duration of computer use, hormonal disturbances, and use of hand sanitisers. An AI model was then built to look at risk factors that could impact disease outcome. The area-under-curve (AUC), sensitivity (se), specificity (sp) and accuracy (ac) were evaluated.

Results

2 AI models were used – a tomographic AI model to classify eyes as “progression” and “stable”, and a second AI model built to evaluate clinical risk factors in each group. About 76% of the cases that were classified as "progression" (from tomographic changes) were correctly predicted as "progression" and 65% of cases that were classified as "stable” were predicted as "no progression", based on changes in clinical risk factors. IgE had the highest information gain, followed by presence of systemic allergies, Vitamin D and eye rubbing, The clinical risk factors model had an ac,se, sp  of, 71.3%, 63% and 87.3%, respectively. 

Conclusions

On the basis of AI, this study demonstrates the importance of using demographic parameters for risk stratification and profiling of patients. This could impact disease progression in keratoconus, thus helping us in screening, diagnosis and timely/appropriate management of these patients.