Deep learning in ectasia detection: corneal tomographic enantiomorphism study

Session Details

Session Title: Anterior Segment Imaging I

Session Date/Time: Monday 24/09/2018 | 16:30-18:00

Paper Time: 17:21

Venue: Room A4

First Author: : B.Lopes UK

Co Author(s): :    R. Vinciguerra   P. Vinciguerra   J. Mello   R. Ambrosio           

Abstract Details

Purpose:

To use machine learning model for detecting clinical ectasia and ectasia risk based on inter-eye asymmetry of tomographic parameters.

Setting:

Rio de Janeiro Corneal Tomography and Biomechanics Study Group

Methods:

Corneal tomography exam data from 2,753 patients was retrospective evaluated. The disease group was composed of 907 patients with diagnosis of keratoconus (KC) from two different centres and the pre-operative state of 15 patients that develop post-LASIK ectasia with normal tomography including a BAD-D (v3) lower than 1.6 in both eyes (PLE-NTm). The control group included 1,831 stable LASIK cases (minimum follow-up of 3 years) from two different centres. The absolute value of the differences between the eyes of multiple tomographic parameters was used to train and to test the artificial intelligence model for optimizing separation among the groups.

Results:

The model with highest accuracy, could correctly classify 95% of the keratoconus cases, and 98% of the stable LASIK cases. Three (20%) PLE-NTm cases were detected. On the external set of validation the model could correctly classify 95% of the cases.

Conclusions:

The deep learning model with the inter-eye asymmetry of tomographic parameters can be used to detect clinical keratoconus with high accuracy. The sensitivity to detect ectasia risk (susceptibility) on the preoperatively of normal tomography PLE cases was limited. However, it represented an improvement in the current screening evaluation. The machine learning model can be used in adjunct to other parameters to further optimize accuracy to detect keratoconus and for screening refractive surgery candidates.

Financial Disclosure:

... receives consulting fees, retainer, or contract payments from a competing company, ... research is funded, fully or partially, by a competing company