Design of algorithms for the construction of 3-dimensional models of the human crystalline lens

Abstract

Quantification of the anterior segment of the human eye (cornea, iris, and crystalline lens) is crucial for the diagnosis and treatment of common eye conditions such as presbyopia and cataracts. Quantification requires measurement of patient’s eye and the construction of accurate 3-D models from these measurements. Measurements are usually obtained using a non-invasive high-resolution optical imaging technique known as Optical Coherence Tomography (OCT). From these measurements, one of the most relevant steps to obtain accurate 3-D models is the development of high-quality and reliable image segmentation algorithms. In this bachelor’s thesis we first describe state-of-the-art image segmentation algorithms of the anterior segment of the eye, which are based on traditional image processing techniques. Also, we present a discussion about their results and shortcomings. Then, we present the main goal of this project: to improve the performance and address the shortcomings of the traditional image processing-based segmentation algorithms. To do so, an image segmentation model based on deep learning was created. A comparison between both image segmentation models will be carried out. The results obtained show that model performance and reliability, key aspects for the quantification of anterior segment of the human eye, are superior using deep learning. The proposed deep learning model could have a strong positive impact on the quantification of the human eye, which in turn could help in the diagnosis and treatment of high-prevalence eye conditions.