Akadēmiskais centrs
Tēma: Konvolucionālā neironu tīkla izmantošana optometrijā, lai noteiktu diabētiskās retinopātijas un tuvredzības pazīmes no tīklenes dibena attēliem bez kodēšanas pieredzes.
Theme: Use of a Convolutional Neural Network in Optometry to detect features of Diabetic Retinopathy and Myopia from retinal fundus images without coding expertise.
The doctoral thesis explores the application of artificial intelligence in optometry and ophthalmology for the early and objective diagnosis of ocular diseases, such as diabetic retinopathy (DR), and myopia. Diabetic retinopathy, for instance, is a progressive ocular disease that involves the appearance of typical signs such as microaneurysms, hemorrhages and exudates visible in the retinal fundus images. Computer vision and machine learning techniques can identify and map these signs for improved and accurate diagnosis and follow-up. However, developing such systems requires transversal skills that eye-care professionals only sometimes possess, limiting the use and diffusion of such tools. This doctoral work aims to evaluate the performance and application of an online-based convolutional neural network (CNN) trained to find the signs of DR without programming and coding skills. Thanks to the application's programmable interface, we could efficiently perform lesion segmentation and detection, opening a route for coding-independent software development. Furthermore, the same neural network was trained to classify the ocular fundus images of myopic and non-myopic patients to detect the presence of this eye condition accurately. The potential of myopia detection in its early stages from a retinal fundus picture to revolutionize the paradigm of refractive error screening, especially in children, is an exciting prospect.
Priekšaizstāvēšana notiks 05.11.2024 plkst. 14.30 attālināti MS Teams, pietekšanās rakstot epastu uz optometrija@lu.lv
Priekšaizstāvēšana notiks angļu valodā.
