Using convolutional neural networks for corneal arcus detection towards familial hypercholesterolemia screening

Abstract

Familial hypercholesterolemia (FH) is a highly undiagnosed disease. Among FH patients, the onset of premature coronary artery disease is 13 times higher than in the general population. Early diagnosis and treatment is essential to prevent cardiovascular diseases and their complications, and to prolong life. One of the clinical criteria of FH is the occurrence of a corneal arcus (CA) among patients, especially those under 45 years old. Therefore, by detecting a CA, it might be possible to reduce the number of undiagnosed FH cases. In this paper, we propose using convolutional neural networks (CNN) for automatic recognition of the presence of a corneal arcus. To achieve this goal, we created a dataset of images of irises containing different stages of CA as well as irises without a CA. The core of the dataset consists of images acquired from patients with a corneal arcus, enroled in the National Centre of Familial Hypercholesterolemia in Gdansk. To increase the number of images, the dataset was complemented with images downloaded from the Internet. This dataset created for training and testing the model consisted of nearly 4000 images. To detect a CA in photographic images, we tested neural network models based on the VGG16, ResNet and Inception architectures. Finally, the performance of the models was evaluated on a set of images acquired from volunteers with a custom mobile application. The accuracy of CA detection in a real life scenario was 88% and the F1 score was 86%