EI Compendex Source List（2022年1月）
EI Compendex Source List（2020年1月）
EI Compendex Source List（2019年5月）
EI Compendex Source List（2018年9月）
EI Compendex Source List（2018年5月）
EI Compendex Source List（2018年1月）
中国科学引文数据库来源期刊列
CSSCI(2017-2018)及扩展期刊目录
2017年4月7日EI检索目录（最新）
2017年3月EI检索目录
最新公布北大中文核心期刊目录
SCI期刊（含影响因子）
论文范文
Abstract：Entropy images, representing the complexity of original fundus photographs, may strengthen the contrast between diabetic retinopathy (DR) lesions and unaffected areas. The aim of this study is to compare the detection performance for severe DR between original fundus photographs and entropy images by deep learning. A sample of 21,123 interpretable fundus photographs obtained from a publicly available data set was expanded to 33,000 images by rotating and flipping. All photographs were transformed into entropy images using block size 9 and downsized to a standard resolution of 100 × 100 pixels. The stages of DR are classified into 5 grades based on the International Clinical Diabetic Retinopathy Disease Severity Scale: Grade 0 (no DR), Grade 1 (mild nonproliferative DR), Grade 2 (moderate nonproliferative DR), Grade 3 (severe nonproliferative DR), and Grade 4 (proliferative DR). Of these 33,000 photographs, 30,000 images were randomly selected as the training set, and the remaining 3,000 images were used as the testing set. Both the original fundus photographs and the entropy images were used as the inputs of convolutional neural network (CNN), and the results of detecting referable DR (Grades 2–4) as the outputs from the two data sets were compared. The detection accuracy, sensitivity, and specificity of using the original fundus photographs data set were 81.80%, 68.36%, 89.87%, respectively, for the entropy images data set, and the figures significantly increased to 86.10%, 73.24%, and 93.81%, respectively (all values <0.001). The entropy image quantifies the amount of information in the fundus photograph and efficiently accelerates the generating of feature maps in the CNN. The research results draw the conclusion that transformed entropy imaging of fundus photographs can increase the machinery detection accuracy, sensitivity, and specificity of referable DR for the deep learning-based system.
1. Introduction
Diabetic retinopathy (DR) is one of the microvascular complications related to diabetes mellitus and a major cause of blindness globally. In the United States, the DR prevalence among diabetic patients is between 20% and 30% [1, 2]. Fundus photography is a direct visual screening tool used to detect DR and has been widely accepted worldwide. However, the detection of DR and assessment of its severity require specialized expertise, and the agreement of interpretation results between examiners varied substantially based on previous studies [3–5]. In addition, many diabetic patients do not have access to effective screening programs and some cannot afford the cost of an ophthalmologist visit [6]. Deep learning is a subset of machine learning, and in modern medicine, using deep learning in fundus photography has emerged as a cost-effective and practical method for automated grading of DR [5, 7, 8]. To implement deep learning in fundus photography for DR grading, a large data set of fundus photography is required, and the amount of data for each grade is preferred to be evenly distributed. However, the retinal images collected from different eye clinics are not standardized (e.g., differences in contrast, brightness, and file size) in all known open web data sets and epidemiology reports [8–10]. Therefore, it is important to preprocess the original images to increase the heterogeneity before putting the images into the training for the automated detection of DR. Several preprocessing methods have been applied to the deep learning of fundus photography [5, 11, 12]. Data augmentation would generate training images utilizing different processing skills or a combination of these skills, such as rotating, shifting, and flipping to the original training images. In addition, contrast and brightness adjustment would increase the heterogeneity and accuracy of the testing, and therefore enhance the automated grading performance. These currently available preprocessing methods can efficiently increase the number of original fundus photographs used for deep learning, without changing the morphology of most predictive features. As the original retinal images are preprocessed or translated to another form of quantitative bioinformative images, they are in essence considered new images by the CNN, so the performance of deep learning will be improved. 
|
|
|
