Optimised artificial intelligence for integrating Scheimpflug tomography and biomechanics for enhanced ectasia detection

Session Details

Session Title: Keratoconus

Session Date/Time: Monday 16/09/2019 | 14:00-16:00

Paper Time: 15:00

Venue: Free Paper Forum: Podium 2

First Author: : R.Ambrosio Jr BRAZIL

Co Author(s): :                                 

Abstract Details

Purpose:

To test, validate and further optimize the Tomographic and Biomechanical Index (TBI), which combines Scheimpflug-based corneal tomography and biomechanics for an enhanced ectasia detection.

Setting:

The Rio de Janeiro Corneal Tomography and Biomechanics Study Group & the BrAIN (Brazilian Study Group of Artificial Intelligence and Corneal Analysis) along with twenty-five international centers.

Methods:

A multicentric retrospective study comprehensively analyzed tomographic and biomechanical data from the Oculus Pentacam and Corvis ST. One eye randomly selected from 1,512 patients with normal corneas, from the preoperative of 238 stable LASIK cases (minimal follow up of two years), and from 1,241 patients with clinical mild to moderate keratoconus were included. The study also included the eyes with normal topography from 496 patients with very asymmetric ectasia (VAE-NT) and the 436 unoperated ectatic (VAE-E) fellow eyes. The current TBI was tested and further optimization of artificial intelligence has been applied for attempting augmenting accuracy.

Results:

For clinical ectasia, the TBI and PRFI (Pentacam Random Forest Index) had AUC (area under the receiver operating characteristics [ROC] curve) of 0.999, being statistically that the AUC of CBI (Corvis Biomechanical Index; 0.961) and BAD-D (Belin/Ambrosio Deviation; 0.992). The AUC of the TBI (0.908) was higher than the PRFI (0.891) for detecting VAE-NT (De Long; p<0.05). Further optimization of the artificial intelligence for integrating the data was achieved with random forest and neural network with leave one out cross-validation, leading the AUC of 0.943 and 0.952 for detecting VAE-NT while maintaining same AUC for clinical ectasia.

Conclusions:

Corneal topography provides excellent accuracy for detecting clinical ectasia. The integration of biomechanical data significantly improves the ability to detect subclinical ectatic disease, such as the VAE-NT. While the current TBI had good accuracy for detecting clinical ectasia, there was a reduction in sensitivity for the VAE-NT cases compared to the original studies. Further improvements in artificial intelligence algorithms to integrate tomography and biomechanics are possible, augmenting the accuracy to detect ectasia and possibly to epitomize the characterization of ectasia susceptibility.

Financial Disclosure:

gains financially from product or procedure presented, travel has been funded, fully or partially, by a competing company, research is funded, fully or partially, by a competing company