Free Papers

Probability Indices to Classify Keratoconus using Artificial Intelligence (PICK-AI): a tool to predict progression in keratoconus (KC)

Free Paper Details

First Author: P.Thakur INDIA

Co Author(s):    R. Shetty   A. Sinha Roy   G. Kundu   R. Narasimhan           

Abstract Details

Purpose:

To study performance of quantitative parameters & indices in determining the progression of KC using AI.

Setting:

Narayana Nethralaya Eye Hospital 121/C ,Chord road,1st R block Rajajinagar , Bengaluru , Karnataka - 560010

Methods:

Over 3500 treatment naïve scans of 206 eyes having keratoconus with good quality score after checking edge detection were exported from Pentacam HR & classified into 2 groups: Stable &Progressing based on Kmax. Any change in Kmax of 1.25 D between two visits of minimum 6 months duration was defined as progression. Keratometry parameters, derived KC indices, & Zernike wavefront aberrations, from both anterior& posterior cornea were given as features to the AI.

Results:

Random forest classifier-based AI model predicted disease progression with area under the curve (AUC) at 0.92, sensitivity& specificity at 0.89& 0.92 respectively. Amongst the features ,detected by the RF classifier-RMS of lower & higher order aberrations of the anterior surface, Steep keratometry of the front surface, index of height decentration (IHD)& index of surface variance (ISV) had higher gain ratios .Using a confusion matrix in a decision tree classifier we were able to accurately predict 81.5% of progressors and 92.8% of stable patients.

Conclusions:

With a combination of keratometry, derived indices & Zernike aberrations, AI model was able to predict disease progression with good accuracy.

Financial Disclosure:

-