loading page

A Deep Learning Analysis Framework for Ophthalmic Diseases and Physical Health from Binocular Fundus Image Pairs
  • +3
  • Dongsheng Zhu,
  • Aiming Ge,
  • Xindi Chen,
  • Shuo Liu,
  • Qiuyang Wang,
  • Jiangbo Wu
Dongsheng Zhu
Fudan University
Author Profile
Aiming Ge
Fudan University

Corresponding Author:[email protected]

Author Profile
Xindi Chen
Fudan University
Author Profile
Shuo Liu
Fudan University
Author Profile
Qiuyang Wang
Fudan University
Author Profile
Jiangbo Wu
Fudan University
Author Profile

Abstract

Some common physical diseases, such as diabetes and hypertension, require comprehensive diagnosis with ophthalmology in clinical practice. However, the huge number of patients has led to insufficient medical supply in many underdeveloped areas. Therefore, this work proposes a novel deep learning framework based on binocular fundus image pairs for the detection of 8 diseases related to the eyes and body. For the first time, the framework improves its basic capabilities by pre-training external fundus datasets and then uses the Multi-scale Dilated Convolution (MDC) module to fuse binocular features to realize the binocular diagnosis. To verify the effectiveness of the method, 4 different convolutional neural network (CNN) architectures are selected as the backbone and tested on the multi-label classification benchmark. Through extensive experimental comparisons, it is found that the innovations proposed in this paper improve the performance of all CNN architectures and achieve state-of-the-art results in the field.