Detection of Keratoconus Progression Using Machine Learning

Keratoconus (KC) is a bilateral ocular disease characterized by progressive thinning and steepening of the cornea, usually in its inferotemporal region. The disease often occurs asymmetrically as one eye is more severely affected by the condition. The changes underlying KC lead to the generation of irregular astigmatism resulting in diminished visual acuity of the patients and can even lead to axial corneal scarring in advanced stages. The disease usually occurs in the second or third decade of life, but can develop at any age. KC is a complex condition involving environmental factors such as age, eye rubbing, contact lens use, atopy, sun exposure, hormones, toxins, as well as a genetic component. However, how these factors contribute to the disease is still unknown and intraindividual differences might exist.

KC can be categorized into different forms based on the stage of the disease. In clinical KC, there are both topographic and slit lamp findings of the disease.

The importance of corneal epithelial imaging in the diagnosis of keratoconus has been further demonstrated in several clinical studies. As new anterior segment optical coherence tomography (AS-OCT) devices provide more detailed measurements for instance of the corneal epithelium. This layer could therefore be an interesting marker for the prediction of KC progression and contribute to earlier diagnosis as well as better outcome of the disease.

The aim of this retrospective study is therefore to determine whether different topographical and volumetric data, for instance epithelial thickness maps (ETM), can be reliably used to predict the progression of KC using a machine learning algorithm.