Screening for retinal diseases is crucial in preventing vision loss. Current protocols rely on manual examinations, limiting large-scale screening, particularly in low-resource areas. Artificial intelligence (AI) shows promise in automating retinal image analysis. Deep learning (DL) systems have demonstrated performance comparable to experts in detecting prevalent diseases like diabetic retinopathy and age-related macular degeneration. However, only a few AI systems have regulatory approval to be used in real-world settings.
This thesis addresses factors hindering AI integration in ophthalmic practice. We investigate the reliability of commercially-available software for automated retinal disease screening, examine the explainability of DL systems' decisions and its impact on trust and clinical use, and evaluate the robustness of DL systems against malicious attacks. Lastly, we explore methods to generate trustworthy AI, a crucial step to bringing the benefits of AI closer to healthcare end-users.
Cristina González Gonzalo (1994) holds a BSc in Audiovisual Systems Engineering from Universidad Carlos III de Madrid and an MSc in Biomedical Engineering from Universidad Politécnica de Madrid. In 2017, she started her PhD at the Diagnostic Image Analysis Group at Radboudumc. Her research interests focus on bringing the potential benefits of AI closer to the final users in healthcare.