Abstract and Transferred Title Page

## English Translation of the English Abstract (Original S. 195) ### Introduction Keratoconus is a common condition; its early detection is very important. It is carried out through careful assessment of topographic images to distinguish between keratoconic, suspect, and normal corneas. Many studies have attempted to find criteria and indicators that support early detection, with varying accuracy metrics. However, the previous approaches relied on a very small portion of the topographic information. Following the qualitative leap in computer science, particularly in the field of artificial intelligence and artificial neural networks, and the emergence of medical applications for assessing and analysing medical images, it was necessary to investigate this technique for detecting suspect and keratoconic corneas. ### Study Objective Investigation of the performance of computer vision technologies and deep learning algorithms for differentiating topographic map images of normal, keratoconic, and suspect corneas. ### Materials and Methods The study consists of two parts. The first part is a retrospective review of patient records and images to compile a training sample of 987 eyes (300 keratoconic, 610 normal, and 77 suspect eyes). The second part is a cross-sectional study to compile the test sample of 422 eyes (13 keratoconic, 366 normal, and 43 suspect eyes). The AI system consists of ten artificial neural networks. Each network is responsible for assessing one map (anterior and posterior tangential maps, anterior and posterior sagittal maps, anterior and posterior elevation maps, anterior, posterior, and equivalent refractive power maps, and pachymetry) and predicting the correct class. The outputs of these networks form the inputs of a final neural network that makes the system's final decision. ### Results The AI system achieved test group accuracy and overall weighted F1-score values of 92.2%–91.2%. No statistically significant difference existed between it, the SIRIUS device software, and the physician (p > 5%). The physician achieved the best result with AI system support compared with all other models: 96.2% accuracy and 95.9% weighted F1-score; the difference was statistically significant (p < 5%). ### Conclusions We recommend using the AI system as a support instrument for the physician in assessing topographic maps. **Keywords:** Keratoconus; artificial intelligence; deep learning; neural networks. ## Transferred Title Page (Original S. 196) **Syrian Arab Republic** **University of Damascus** **Faculty of Medicine** **Department of Ophthalmology** ### Application of Computer Vision Techniques and Deep Learning Algorithms for Differentiation of Topographic Images of Normal, Keratoconic, and Suspect Corneas A dissertation submitted in partial fulfilment of the requirements for the degree of Doctor of Philosophy (PhD) in Ophthalmology **Author:** Dr. Mohammad Zafrallah Askar **Supervisor:** Yosra Haddeh, Professor of Ophthalmology at the Faculty of Medicine, University of Damascus **Co-supervisor:** Madhat Alsoos, Department of Artificial Intelligence, Faculty of Informatics and Information Technology, University of Damascus **Study Year:** 2020–2021