Automated geographic atrophy segmentation for SD-OCT images based on two-stage learning model

Label embedding has been widely used as a method to exploit label dependency with dimension reduction in multilabel classification tasks. However, existing embedding methods intend to extract label correlations directly, and thus they might be easily trapped by complex label hierarchies. To tackle this issue, we propose a novel Two-Stage Label Embedding (TSLE) paradigm that involves Neural Factorization Machine (NFM) to jointly project features and labels into a latent space. In encoding phase, we introduce a Twin Encoding Network (TEN) that digs out pairwise feature and label interactions in the first stage and then efficiently learn higherorder correlations with deep neural networks (DNNs) in the second stage. After the codewords are obtained, a set of hidden layers is applied to recover the output labels in decoding phase. Moreover, we develop a novel learning model by leveraging a max margin encoding loss and a label-correlation aware decoding loss, and we adopt the mini-batch Adam to optimize our learning model. Lastly, we also provide a kernel insight to better understand our proposed TSLE. Extensive experiments on various real-world datasets demonstrate that our proposed model significantly outperforms other state-ofthe-art approaches.

Download Full-text

Origins of Ghost Drusen: A follow-up Analysis

Ophthalmology Point of Care ◽

10.5301/oapoc.0000011 ◽

2017 ◽

Vol 1 (1) ◽

pp. oapoc.0000011

Author(s):

Caroline Bottin ◽

Olivia Zambrowski ◽

Giuseppe Querques ◽

Salomon Yves Cohen ◽

Mayer Srour ◽

...

Keyword(s):

Pigment Epithelium ◽

Retinal Pigment ◽

Geographic Atrophy ◽

Primary Lesion ◽

Age Related Macular Degeneration ◽

Age Related ◽

Soft Drusen ◽

Initial Lesions ◽

Sd Oct

Purpose Ghost drusen (GD) are pyramidal or dome-shaped retinal pigment epithelium elevations observed in some geographic atrophy (GA) areas in the context of age-related macular degeneration (AMD). The purpose was to investigate the first morphologic features preceding GD on spectral-domain optical coherence tomography (SD-OCT) on patients with GA associated with AMD. Methods A retrospective observational study was performed on a series of patients with GA that had at least 3 years of follow-up. Using the follow-up tool of SD-OCT, we tracked the initial lesions that could lead to GD. Results Among 442 patients with GA, 37 had well defined GD (8%). We included the 17/37 patients (31 eyes) with at least 3 years of follow-up for analysis, which led to a total of 582 counted GD. Most GD were already present at the first visit, and remained stable. However, on 13 of the 582 analyzed GD (2.2%), soft drusen were shown as the initial lesion, which progressively turned into GD. Conclusions GD were observed in less than 10% of eyes with GA. None of the ghost drusen turned into another shaped lesion, suggesting that GD is a possible final stage of evolution. In a few cases, large drusen were shown as the primary lesion that progressed into GD.

Download Full-text

Early Detection of Incipient Retinal Pigment Epithelium Atrophy Overlying Drusen with Fundus Autofluorescence vs. Spectral Domain Optical Coherence Tomography

Journal of Ophthalmology ◽

10.1155/2020/9457457 ◽

2020 ◽

Vol 2020 ◽

pp. 1-8 ◽

Cited By ~ 1

Author(s):

Anabel Rodríguez ◽

Marc Biarnés ◽

Rosa M. Coco-Martin ◽

Anna Sala-Puigdollers ◽

Jordi Monés

Keyword(s):

Optical Coherence Tomography ◽

Retinal Pigment Epithelium ◽

Median Time ◽

Pigment Epithelium ◽

Fundus Autofluorescence ◽

Retinal Pigment ◽

Geographic Atrophy ◽

Optical Coherence ◽

Rpe Atrophy ◽

Sd Oct

Purpose. This study aims to find out which tool, fundus autofluorescence (FAF) or spectral domain optical coherence tomography (SD-OCT), is more sensitive in detecting retinal pigment epithelium (RPE) demise overlying drusen and can, therefore, help predict geographic atrophy (GA) appearance in Age-Related Macular Degeneration (AMD). Methods. A single-site, retrospective, observational, longitudinal study was conducted. Patients with intermediate AMD (iAMD) (large (>125 μm) or intermediate (63–125 μm) drusen with hyper/hypopigmentation) with a minimum follow-up of 18 months were included. Drusen with overlying incipient RPE atrophy were identified on SD-OCT defined as choroidal hypertransmission or nascent geographic atrophy (nGA). These selected drusen were, then, traced backwards in time to determine if incipient RPE atrophy overlying drusen was observed on FAF (well-demarcated region of absence of autofluorescence) before, simultaneously, or after having detected the first signs of incipient RPE atrophy on SD-OCT. The number of drusen in which signs of incipient RPE atrophy was detected earlier using FAF or SD-OCT was compared. The time elapsed from the identification with the more sensitive method to the other was recorded and analyzed. Results. One hundred and thirty-three drusen in 22 eyes of 22 patients were included. Of these, 112 (84.2%) drusen showed choroidal hypertransmission and 21(15.8%) nGA. Early signs of atrophy overlying drusen were found simultaneously on SD-OCT and FAF in 52 cases (39.1%, 95% CI 30.8–47.9%), earliest on FAF in 51 (38.3%, 95% CI 30.0–47.2%) and first on SD-OCT in 30 (22.6%, 95% CI 15.8–30.6%; p<0.05). Statistically significant differences were found between both techniques (p=0.005), with FAF detecting it earlier than SD-OCT. When RPE atrophy was found first on FAF, the median time to diagnosis with SD-OCT was 6.6 months (95% CI 5.5 to 8.6), while if detection occurred earlier on SD-OCT, the median time until identification with FAF was 12.6 months (95% CI 6.0 to 23.4; p=0.0003). Conclusions. In iAMD cases in which early atrophy overlying drusen is not detected simultaneously in FAF and SD-OCT, FAF was significantly more sensitive. Nevertheless, a multimodal approach is recommended and required to evaluate these patients.

Download Full-text