Decision support system for an early-stage keratoconus diagnosis

Abstract

Currently, there is a wide variety of different diseases that exist, a lot of which can be hardly prompt diagnosed even by medical specialists. This paper presents a method for early-stage diagnosis of ophthalmologic disorder keratoconus. Working with medical imagery that was captured by a rotating Scheimpflug camera system for anterior segment analysis, the goal was to create a decision support system to aid ophthalmologists in prompt detection of the disorder to eliminate the chance of further surgical intervention. Given approach uses several steps to achieve that goal, such as find the region of interest on the medical imagery using a Single Shot MultiBox Detector, filter and binarize the image, locate the cornea and approximate the curve with least squares fitting, classify the stage according to the previously labeled dataset by medical specialists with a random forest method. The suggested approach achieves 76% precision on a dataset containing 500 images of the patients with the first stage of keratoconus and healthy patients. The final result was compared with modern existing medical methods that are usually used in ophthalmologic clinics by medical specialists.