Automated segmentation of geographic atrophy of the retinal epithelium via random forests in AREDS color fundus images

Abstract

Background-Age-related macular degeneration (AMD), left untreated, is the leading cause of vision loss in people older than 55. Severe central vision loss occurs in the advanced stage of the disease, characterized by either the in growth of choroidal neovascularization (CNV), termed the "wet" form, or by geographic atrophy (GA) of the retinal pigment epithelium (RPE) involving the center of the macula, termed the "dry" form. Tracking the change in GA area over time is important since it allows for the characterization of the effectiveness of GA treatments. Tracking GA evolution can be achieved by physicians performing manual delineation of GA area on retinal fundus images. However, manual GA delineation is time-consuming and subject to inter-and intraobserver variability.Methods-We have developed a fully automated GA segmentation algorithm in color fundus images that uses a supervised machine learning approach employing a random forest classifier. This algorithm is developed and tested using a dataset of images from the NIH-sponsored Age Related Eye Disease Study (AREDS). GA segmentation output was compared against a manual delineation by a retina specialist.Results-Using 143 color fundus images from 55 different patient eyes, our algorithm achieved PPV of 0.82±0.19, and NPV of 0:95±0.07.Discussion-This is the first study, to our knowledge, applying machine learning methods to GA segmentation on color fundus images and using AREDS imagery for testing. These HHS Public Access

Comparing humans and deep learning performance for grading AMD: A study in using universal deep features and transfer learning for automated AMD analysis

“…Finally, the three feature vectors are concatenated together to form the feature vector used for classification. We note here that this approach is motivated by the fact that, for color fundus images, features near the center of the macula are of greater significance for classifying the AMD severity category[10][14]. …”

“…In particular, prior to classification, conventional methods have included a step in which specific visual features are computed such as wavelets or SIFT features [9],[14],[15]. This manual selection (or feature design) can result in a set of features that are suboptimal and overly specialized to a specific data set resulting in poor generalization to larger or unknown data sets.…”

Detection of age-related macular degeneration via deep learning

“…It has been observed previously that features near the center of the macula tend to be more important for classifying the AMD severity category. 11 To account for this fact, our approach uses different concentric square image grid areas (windows) within the fundus image and concatenates the results into a single feature vector. Figure 2 shows the grid areas considered.…”

“…In particular, prior to classification, conventional methods include a step in which specific visual features, such as wavelets, SURF, or SIFT, are computed. [9][10][11] This manual selection (or feature design) can result in a set of features that is overly specialized to a specific data set resulting in reduced applicability to larger or unknown data sets. By contrast, DCNN's features are not manually designed but are found during network weight training and optimization.…”

Effect of ARMS2 gene polymorphism on intravitreal ranibizumab treatment for neovascular age-related macular degeneration

“…The two main types of AMD are dry type AMD and neovascular type AMD. Neovascular AMD is characterized by the invasion of subretinal pigment epithelium and subretinal spaces by choroidal neovascularization, and geographic atrophy is typified by the degeneration of the choriocapillaris, Bruch's membrane, retinal pigment epithelium, and retina (Danis et al, 2015;Feeny et al, 2015;Schütze et al, 2015;Ferrington et al, 2016).…”