Detection of early keratoconus through machine-learning and implementation in the Pentacam as a screening tool

Date: 29 November 2016

Venue: UAntwerp, Campus Drie Eiken, Building Q, aula F. Nédée - Universiteitsplein 1 - 2610 Wilrijk (route: UAntwerpen, Campus Drie Eiken)

Time: 5:30 PM

PhD candidate: Irene Ruiz Hidalgo

Principal investigator: Prof M.-J. Tassignon & Prof J. Rozema

Short description: PhD defence Irene Ruiz Hidalgo - Faculty of Medicine and Health Sciences


Keratoconus is a corneal disorder in which the cornea assumes a conical shape as a result of thinning and protusion. The deformation is progressive, bilateral and asymmetric, resulting in loss of visual acuity. Although the underlying cause of the disease is still unknown, the general assumption is that it is associated with a localized loss in corneal elasticity. Keratoconus is the third leading cause of corneal transplant in developed countries, which is indicated in case corneal scarring appears. Treatment options are largely symptomatic and range from contact lenses, ring segments and corneal cross-linking for mild cases to corneal transplant for the advanced ones.

The main goal of this PhD thesis was to develop a screening tool to help clinicians with the detection and management of keratoconus in different stages in a clinical environment, with a special focus on the early forms of the disease such as forme fruste keratoconus. This is of vital importance to halt the progressive loss in visual quality by performing corneal cross-linking on patients affected by it.

Detecting moderate keratoconus is not challenging since it can easily be seen through routine clinical tests. However, earlier cases such as subclinical or forme fruste keratoconus have always been a challenge for ophthalmologists, especially when the typical clinical signs and symptoms that differentiate keratoconus from normal corneas cannot be observed. In the pas years, several topographic indices have been developed with this goal; however, most of them show difficulties differentiating between subclinical keratoconus and normal corneas, due to the topographical similarities between them.

This project introduces a Pentacam-based artificial intelligence tool, the Keratoconus Assistant (KA) that is based in a combination of corneal keratometry, pachymetry and elevation parameters. This tool is an open-source software that can be installed in any Pentacam computer as an add-on program to assist with the differential diagnosis of patients with keratoconus. The outcome is reported by a pop-up message on the screen with the probability of the patient measured of having the disease.

The KA shows an improved accuracy detecting forme fruste keratoconus without clinical signs with respect to the existing methods and it is of practical benefit for the prevention of post-refractive ectasia in corneal refractive surgery.

Entrance fee: free