Seeing Is Misbelieving: A Computational and Statistical Framework for Characterizing Visual Distortion in Central Vision Loss

Abstract

Diseases affecting the central visual field, including macular telangiectasia type 2 (MacTel), Stargardt macular dystrophy, and age-related macular degeneration, are a leading cause of blindness globally and significantly diminish quality of life. While advanced imaging tools like OCT provide anatomical insight into the physical basis of these diseases, they often overlook a critical component: the patient's subjective visual experience. Visual field distortion, where straight lines appear wavy or portions of images appear missing, is a common symptom that clinicians assess using the Amsler grid—a simple but powerful paper test. The Amsler grid test is a standard visual distortion screening practice and often complements high-resolution retinal imaging tests. Despite its widespread use, the Amsler grid relies on qualitative interpretation and verbal patient reports, which can be inconsistent and imprecise—in fact, subjective patient reporting often leads clinicians to misdiagnose macular disease or miss it altogether. There is a need for a quantitative framework to analyze patient-annotated Amsler grids. In this thesis, we propose a novel computational and statistical framework to characterize visual distortions observed on patient-annotated Amsler grids. We applied this framework to a patient with MacTel and found that our framework effectively captured variation within eyes and asymmetries between eyes. By demonstrating the feasibility of drawing clinically relevant insights from hand-drawn patient data, we show that our framework not only has the potential to enhance diagnostic sensitivity but also bridges the gap between objective imaging and the subjective patient experience. Ultimately, our framework could reshape the current landscape of retinal disease monitoring and diagnosis.