- Vienna '18
- Athens 2019
- ESCRS Player
- On Demand
- ESCRS iLearn
- ESCRS YO's
Session Title: Imaging II
Session Date/Time: Monday 07/10/2013 | 16:30-18:00
Paper Time: 16:30
Venue: Main Lecture Hall (Ground Floor)
First Author: : D.Smadja FRANCE
Co Author(s): :
To develop a method for automatizing the detection of subclinical keratoconus based on a tree classification
University Hospital of Bordeaux, France
A total of 372 eyes of 197 patients were prospectively enrolled and imaged with a Dual Scheimpflug Analyzer. Corneas were classified into 3 conditions: 177 normal eyes of 95 subjects, 47 eyes of 47 patients with form fruste keratoconus (FFKC) and 148 eyes of 102 patients with keratoconus. 56 parameters derived from both anterior and posterior corneal measurements were analyzed for each eye and a machine learning algorithm, the Classification And Regression Tree (CART) was used to classify the eyes into the 3 above-mentioned conditions. The error of the decision tree was estimated by a cross validation process.
The new discriminating rule developed with CART allowed to discriminate between normal and keratoconus with 100% sensitivity and 99.5% specificity and between normal and FFKC with 93.6% sensitivity and 97.2% specificity. This discriminating rule selected as the most discriminant variables, parameters that were related to posterior surface asymmetry and thickness spatial distribution. The performance of this new discriminating rule was found significantly better than the Keratoconus Prediction Index (KPI), currently used by the Dual Scheimpflug analyzer, with an area under the receiver operating characteristics of 0.90 with the new rule compared to 0.77 with the KPI (p < 0.001).
The machine learning classifier showed a very good performance for discriminating between normal corneas and FFKC and provided a tool that is closer to an automated medical reasoning. This might help in the surgical decision by helping the clinician in detecting ectasia-susceptible corneas before refractive surgery.
Please wait while information is loading.