ESCRS - FPT02.11 - Kc App : A Novel Artificial Intelligence Based Application For Simplified Guidelines In Keratoconus Management

Kc App : A Novel Artificial Intelligence Based Application For Simplified Guidelines In Keratoconus Management

Published 2022 - 40th Congress of the ESCRS

Reference: FPT02.11 | Type: Free paper | DOI: 10.82333/m7bv-q894

Authors: Savitri Deval* 1 , Rohit Shetty 2 , Pooja Khamar 1 , Gairik Kundu 3

1CATARACT AND REFRACTIVE SURGERY,NARAYANA NETHRALAYA EYE HOSPITAL, BENGALURU, INDIA,BENGALURU,India, 2CATARACT, CORNEA AND REFRACTIVE SERVICES,NARAYANA NETHRALAYA EYE HOSPITAL, BENGALURU, INDIA,BENGALURU,India, 3CORNEA AND REFRACTIVE SERVICES,NARAYANA NETHRALAYA EYE HOSPITAL, BENGALURU, INDIA,BENGALURU,India

Purpose

To study the performance of AI based smart phone application in identifying keratoconus progressors and suggest an appropriate case based management.

Setting

Patients visiting Narayana Nethralaya, Bengaluru on a out-patient basis, who were diagnosed as having Keratocnus were shortlisted based on their scans and recruited in the study after obtaining well informed consent.

Methods

A total of 2500 scans of  200 eyes, having good quality were exported from Pentacam HR and were then classified as stable or progressors based on their Kmax values. Keratometry parameters, KC indices and Zernike wavefront aberrations of these eyes were then subsequently fed into the AI software. Machine learning was used to teach AI algorithm for various management options for keratoconus like Collagen Cross-linking, Intra-stromal Corneal Ring Segments (ICRS), Topography guided Custom Ablation Treatment (TCAT), Deep Lamellar Anterior Keratoplasty (DALK), Penetrating Keratoplasty (PK) and Femtosecond laser assisted Lamellar Keratoplasty. Progression and management derived data was then used to build a smart phone application.

Results

Random forest classifier - based AI model predicted the disease progression with area under the curve at 0.92, along with the sensitivity and specificity of  0.8 and 0.87 respectively. Sensitivity and specificity of the KC app in successfully identifying keratoconus progressors and suggesting a case appropriate treatment was 96.4 %.

Conclusions

KC app is the first ever AI based software for improving the predictability of keratoconus progression and deciding its management. It’s an excellent tool to sort, organize and display relevant information in a simple and accessible manner, serving as a one stop solution to clinical practitioners.