June 2021
Volume 62, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2021
Diagnosing and segmenting choroidal neovascularization in optical coherence tomographic angiography using deep learning
Author Affiliations & Notes
  • JIE WANG
    Oregon Health & Science University Casey Eye Institute, Portland, Oregon, United States
  • Tristan T. Hormel
    Oregon Health & Science University Casey Eye Institute, Portland, Oregon, United States
  • kotaro tsuboi
    Oregon Health & Science University Casey Eye Institute, Portland, Oregon, United States
  • Xiaogang Wang
    Shanxi Eye Hospital, Taiyuan, Shanxi, China
  • Xiaoyan Ding
    Sun Yat-Sen University Zhongshan Ophthalmic Center, Guangzhou, Guangdong, China
  • Xiaoyan Peng
    Beijing Tongren Eye Center, Beijing, China
  • Steven T Bailey
    Oregon Health & Science University Casey Eye Institute, Portland, Oregon, United States
  • Yali Jia
    Oregon Health & Science University Casey Eye Institute, Portland, Oregon, United States
  • Footnotes
    Commercial Relationships   JIE WANG, None; Tristan Hormel, None; kotaro tsuboi, None; Xiaogang Wang, None; Xiaoyan Ding, None; Xiaoyan Peng, None; Steven Bailey, None; Yali Jia, OptoVue (F), OptoVue (P)
  • Footnotes
    Support  National Institutes of Health (R01 EY027833, R01 EY024544, P30 EY010572); Unrestricted Departmental Funding Grant and William & Mary Greve Special Scholar Award from Research to Prevent Blindness (New York, NY)
Investigative Ophthalmology & Visual Science June 2021, Vol.62, 2159. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      JIE WANG, Tristan T. Hormel, kotaro tsuboi, Xiaogang Wang, Xiaoyan Ding, Xiaoyan Peng, Steven T Bailey, Yali Jia; Diagnosing and segmenting choroidal neovascularization in optical coherence tomographic angiography using deep learning. Invest. Ophthalmol. Vis. Sci. 2021;62(8):2159.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Purpose : To diagnose and segment choroidal neovascularization (CNV) in a real-world multi-center clinical optical coherence tomographic angiography (OCTA) dataset.

Methods : In this study, a total of 10125 OCTA scans from 2497 eyes, including 4260 CNV scans and 5865 non-CNV scans, were collected from 5 eye clinics. Selected scans included 3×3-mm and 6×6-mm macular scans. All scans were included regardless of image quality. CNV scans were collected from multiple diseases, including neovascular age-related macular degeneration (AMD), pathological myopia, polypoidal choroidal vasculopathy (PCV), and other rare retinal diseases. The non-CNV dataset consisted of a heterogenous group including heathy controls, non-neovascular AMD, diabetic retinopathy, branch retinal vein/artery occlusion, and central serous chorioretinopathy. Two experts (JW and KT) graded projection resolved OCTA images and manually delineated CNV membrane area using both en face of outer retina and cross-sectional OCTA images. Multiple representations of both en face and volumetric OCT&OCTA were fed into a custom designed hybrid multi-task convolutional neural network (CNN) that produces a CNV diagnosis and membrane segmentation. 5-fold cross-validation was applied to evaluate the performance of the proposed method.

Results : For CNV diagnosis, the sensitivities were 96% and 91% on 3×3-mm and 6×6-mm scans with 95% specificity, respectively. Of all scans with CNV, 2% of CNV scans were incorrect due to segmentation error preventing CNV detection. Of low quality scans (n=993) with a signal strength less than 50, CNV was correctly detected 97.3% of the time. The method was able to accurately diagnose CNV and segment CNV membranes on both 3×3-mm and 6×6-mm in neovascular AMD (Fig.1) and also showed reliable performance on challenging scans in other retinal disease, e.g. PCV, and myopia CNV(Fig.2).

Conclusions : The proposed method can accurately diagnose and segment CNV in a real-world clinical dataset. These results could enable automated CNV screening and quantification in clinic and lead to improved artificial intelligence-aided CNV diagnosis.

This is a 2021 ARVO Annual Meeting abstract.

 

 

×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×