Deep Learning Based Discrimination of Corneal Ulcer Patterns Using Fluorescein Staining Images

Wang，Zhonghua; Huang，Yijin; Lyu，Junyan; Cheng，Pujin; Tang，Xiaoying

10.1145/3451421.3451449

2020-12-05

ACM International Conference Proceeding Series

126-129

In this paper, we proposed an automatic pipeline for classifying three types of corneal ulcer patterns (point like, mixed, flaky). We first made use of H-channel related information in HSV space within cornea to perform corneal ulcers' coarse segmentation. This result was then sequentially classified using two binary classification models: 1) Point-like ulcer against mixed and flaky ulcer, and 2) mixed ulcer against flaky ulcer. The proposed pipeline was evaluated on 712 corneal ulcer images, with an accuracy of 0.858 and a kappa score of 0.796 having been obtained. We also investigated the performance of our proposed pipeline with different baseline network architectures, and identified Resnet101 to be the best-performing one.

Corneal Ulcers Image Classification Neural Networks

第一

英语

EI

20213610870635

Diseases ; Network architecture ; Pipelines

Pipe, Piping and Pipelines:619.1

2-s2.0-85114272405

Scopus

Department of Electrical and Electronic Engineering,Southern University of Science and Technology,Shenzhen,China

Wang，Zhonghua,Huang，Yijin,Lyu，Junyan,et al. Deep Learning Based Discrimination of Corneal Ulcer Patterns Using Fluorescein Staining Images[C],2020:126-129.