Copenhagen 2016 Registration Programme Exhibitor Information Virtual Exhibition Satellite Meetings Glaucoma Day 2016 Hotel Star Alliance

10 - 14 Sept. 2016, Bella Center, Copenhagen, Denmark

This Meeting has been awarded 27 CME credits


escrs app advert yo advert

Neural network to guide keratoconus treatment with ICRS

Search Title by author or title

Session Details

Session Title: Moderated Poster Session: Hot Topics in Refractive Surgery

Session Date/Time: Saturday 10/09/2016 | 14:00-15:00

Paper Time: 14:20

Venue: Poster Village: Pod 1

First Author: : J.Alio SPAIN

Co Author(s): :    P. Sanz Diez   F. Versaci   S. Faini   A. Vega-Estrada     

Abstract Details


To described and validate a neural network aiming to optimized the treatment of keratoconus with ICRS.


Vissum, Alicante, Spain. Division of Ophthalmology, Miguel Hernández University, Alicante, Spain and CSO, Florence, Italy


A neural network created from a large number of cases implanted with the different models available of ICRS. Preoperative and postoperative data: visual and refractive parameters, corneal topography and pachymetry was analyzed. To create a model that correctly maps the input to the output using historical data so that the model can then be used to produce the output when the desired output is unknown. Two groups of patients were compared: a) commercial group (simulation of postoperative results from commercial nomogram) and b) neural network group (simulation of postoperative results from the decision proposed by the neural network).


Fifteen cases with ages ranging from 26 to 49 years (mean age of 37.00±7.69 years) comprised this study. In the commercial group was observed a reduction of flattest keratometry (K1) (49.24±4.66D to 46.26±3.41D), steepest keratometry (K2) (52.98±4.81D to 50.18±4.78D) and mean keratometry (KM) (51.11±4.66D to 48.22±4.02D). In the neural network group was observed a reduction of K1 (49.24±4.66D to 48.14±4.34D) of K2 (52.98±4.81D to 50.76±4.30D) and KM (51.11±4.66D to 49.45±4.25D). No significant differences between the study groups were found in K2 and KM values (p=0.06 and p=0.18, respectively). However, significant differences between groups were found in K1 values (p=0.04).


Neural network analysis correctly proposed an adequate ICRS selection for the treatment of keratoconus. Increasing the input data into a neural network may lead to a more accurate and optimized results.

Financial Disclosure:


Back to previous