Distinguishing retinal angiomatous proliferation from polypoidal choroidal vasculopathy with a deep neural network based on optical coherence tomography

Daniel Duck-Jin Hwang,Joon Seo Hwang,Ji In Park,Junseo Ko,Jeewoo Yoon,Kyu Hyung Park,Seong Choi,Jeong Mo Han,Hak Jun Lee,Jinyoung Han,Joonhong Sohn

doi:10.1038/s41598-021-88543-7

Abstract

This cross-sectional study aimed to build a deep learning model for detecting neovascular age-related macular degeneration (AMD) and to distinguish retinal angiomatous proliferation (RAP) from polypoidal choroidal vasculopathy (PCV) using a convolutional neural network (CNN). Patients from a single tertiary center were enrolled from January 2014 to January 2020. Spectral-domain optical coherence tomography (SD-OCT) images of patients with RAP or PCV and a control group were analyzed with a deep CNN. Sensitivity, specificity, accuracy, and area under the receiver operating characteristic curve (AUROC) were used to evaluate the model’s ability to distinguish RAP from PCV. The performances of the new model, the VGG-16, Resnet-50, Inception, and eight ophthalmologists were compared. A total of 3951 SD-OCT images from 314 participants (229 AMD, 85 normal controls) were analyzed. In distinguishing the PCV and RAP cases, the proposed model showed an accuracy, sensitivity, and specificity of 89.1%, 89.4%, and 88.8%, respectively, with an AUROC of 95.3% (95% CI 0.727–0.852). The proposed model showed better diagnostic performance than VGG-16, Resnet-50, and Inception-V3 and comparable performance with the eight ophthalmologists. The novel model performed well when distinguishing between PCV and RAP. Thus, automated deep learning systems may support ophthalmologists in distinguishing RAP from PCV.

Highlights

This cross-sectional study aimed to build a deep learning model for detecting neovascular agerelated macular degeneration (AMD) and to distinguish retinal angiomatous proliferation (RAP) from polypoidal choroidal vasculopathy (PCV) using a convolutional neural network (CNN)
Neovascular AMD, neovascular age-related macular degeneration; RAP, retinal angiomatous proliferation; PCV, polypoidal choroidal vasculopathy; Optical coherence tomography (OCT), optical coherence tomography. a The sensitivity of the classifier for detecting wet AMD eyes was 99.2%, the specificity was 99.1%, and the accuracy was 99.1%. b The sensitivity of the classifier for detecting RAP was 89.4%, the specificity was 88.8%, and the accuracy was 89.1%
When distinguishing between AMD and normal cases, the proposed model had 99.1% accuracy, which is higher than that of most of the other well-known CNN models; VGG-16, Resnet, and Inception showed 98.4%, 95.1%, and 99.1% accuracy, respectively

Summary

Introduction

This cross-sectional study aimed to build a deep learning model for detecting neovascular agerelated macular degeneration (AMD) and to distinguish retinal angiomatous proliferation (RAP) from polypoidal choroidal vasculopathy (PCV) using a convolutional neural network (CNN). Spectral-domain optical coherence tomography (SD-OCT) images of patients with RAP or PCV and a control group were analyzed with a deep CNN. Diagnosing RAP and PCV with deep learning techniques using only noninvasive OCT is promising. There has been much effort in assessing neovascular AMD using OCT, to the best of our knowledge, no study has reported the use of deep learning techniques to distinguish RAP from PCV. We propose a deep CNN model that uses OCT scans to distinguish RAP from PCV. Using gradient-weighted class activation mapping (Grad-CAM)[23] heat maps, specific features determined by the proposed model are visualized to facilitate the understanding of structural differences between RAP and PCV

Methods

Results

Conclusion