Improving the Understanding of Chronic Central Serous Chorioretinopathy through Multimodal Image Processing and Deep Learning

Central serous chorioretinopathy (CSCR) is an idiopathic eye disease characterized by the accumulation of serous fluids below the retina under involvement of the choroid. To this day, the pathogenesis of CSCR is not well understood and no standardized treatment has been established. Untreated, CSCR can lead to severe vision impairment, including loss of sharpest vision. One challenge associated with the analysis of CSCR is the multitude of different image modalities acquired for CSCR patients. Hence, in an effort to improve the understanding of CSCR, this dissertation proposes a novel retinal image analysis pipeline, consisting of deep learning approaches capable of 1) the automatic extraction of biomarkers from specific image modalities and 2) fusing findings from multiple image modalities via multimodal registration. By applying both together, we are able to perform automated large-scale image analysis of CSCR cohorts. Evaluations on both public datasets and in-house CSCR datasets from the University Eye Clinic of Kiel, Germany, and the Leiden University, Netherlands, indicate that our proposed approaches work well individually as well as when combined in the pipeline.