Speaker :- Vaanathi S
Age-related macular degeneration (AMD), a sight-threatening condition among aged population, is recognized as a leading cause of blindness. Retinal color fundus images, captured by fundus camera, are used to detect the visual signs related to AMD even in pre-symptomatic stages. This research work aims to develop a set of image analysis algorithms for detecting AMD at various stages on retinal color fundus images, which could be integrated to create a system for AMD screening.
We localize the main anatomical structures – the optic disc (OD) and macula as a pre-requisite to specify the region of interest (ROI). We develop algorithms to detect and segment various disease signs related to AMD, namely drusen, geographic atrophy (GA) and choroidal neovascularization (CNV), within the ROI. For differentiating the detected signs from other normal and pathological structures, we propose discriminative features based on contrast and visual appearance information, and apply statistical pattern classification frameworks which assign confidence score to each detected sign. Based on disease sign-level outcomes and their cumulative structural and quantitative information, we build image-level analytics that can be trained for applications such as image-based disease screening.
The key contributions of this work are: (1) a novel integrated approach to automatically localize OD and macula, (2) an adaptive super-candidate based approach for detection and segmentation of drusen of various sizes, (3) structure and appearance based features for discrimination between hard and soft drusen, followed by their quantification, (4) combination of rule based and statistical methods for AMD screening.
The performance of the developed algorithms has been extensively evaluated on publicly available datasets and through comparison with state of the art. The developed algorithms can form the core of a screening/grading system for AMD, to enable early detection, timely diagnosis and treatment.