多算法融合的眼底图像渗出液分割

doi:10.16180/j.cnki.issn1007-7820.2020.11.005

摘要/Abstract

摘要：

利用眼底图像中渗出液的亮度与边缘特征,文中采用一种多算法融合的渗出液自动检测分割方法来解决目前传统算法灵敏度低以及检测中存在视盘和其它微血管瘤等暗病灶的干扰等问题。为了提高分割效率和准确率,文中对原始图像进行顶帽底帽变换来增强图像对比度,采用GA与KSW熵法相结合的双阈值分割法对眼底图像进行渗出液分割。实验在Kaggle数据库上进行测试,结果显示该算法在像素层统计的SE和阳性预测值PPV分别为83.6%和93.2%,在图像层统计的SE、SP与AC分别为95.2%、86.2%和90.8%。在另一个独立的DIARETDB1数据库上进行测试,获得的结果分别为82.4%、93.3%、93.6%、96.2%和89.9%。与其它算法对比,文中方法可以很好地区分开渗出液与暗病灶,且检测时间短,具有准确性和高效性。

关键词: 糖尿病视网膜病变, 眼底图像, 遗传算法, KSW熵, 图像分割, 渗出液

Abstract:

Based on the luminance and edge characteristics of the exudates in fundus images, a multi-algorithm fusion method for automatic detection of the exudate is adopted in this paper to solve the problems of low sensitivity of the traditional algorithm and interference of dark lesions such as optic disc and other microangiomas in the detection results. In order to improve the segmentation efficiency and accuracy, this study uses top-hat and boottom-hat to enhance the image contrast of the original image, and then a dual threshold segmentation method combining genetic algorithm and optimal histogram entropy method is proposed to preliminarily segment the image. The experimental results show that the sensitivity and PPV of the algorithm are 83.6% and 93.2% at the pixel level, and the SE, specificity and accuracy are 95.2%, 86.2% and 90.8% respectively at the image level. The results obtained by testing on another independent DIARETDB1 database are 82.4%, 93.3%, 93.6%, 96.2%, 89.9%. Compared with other algorithms, this method can distinguish the exudates from other dark lesions, and the detection time is short, accurate and efficient.

Key words: diabetic retinopathy, fundus image, genetic algorithm, KSW entropy, image segmentation, hard exudates

中图分类号:

TP391

杨振宇,傅迎华,付东翔,王雅静. 多算法融合的眼底图像渗出液分割[J]. 电子科技, 2020, 33(11): 24-30.

YANG Zhenyu,FU Yinghua,FU Dongxiang,WANG Yajing. HEs Segmentation of Fundus Images by Multi-algorithm Fusion[J]. Electronic Science and Technology, 2020, 33(11): 24-30.

图/表 5

图1

图2

图3

表1

图4

参考文献 30

[1]	Cavan D, Makaroff L, Da Rocha Fernandes J, et al. The diabetic retinopathy barometer study: global perspectives on access to and experiences of diabetic retinopathy screening and treatment[J]. Diabetes Research and Clinical Practice, 2017,129(8):16-24.
[2]	Shah A R, Gardner T W. Diabetic retinopathy: research to clinical practice[J]. Clinical Diabetes and Endocrinology, 2017,3(1):9-16.
[3]	Jyoti P M, Samarendra D. An effective fovea detection and automatic assessment of diabetic maculopathy in color fundus images[J]. Computers in Biology and Medicine, 2016,74(9):30-44.
[4]	Quellec, Gwenolé, Charrière, et al. Deep image mining for diabetic retinopathy screening[J]. Medical Image Analysis, 2017,39(2):178-193.
[5]	Murugeswari S, Sukanesh R. Investigations of severity level measurements for diabetic macular oedema using machine learning algorithms[J]. Irish Journal of Medical Science, 2017,186(4):929-938. doi: 10.1007/s11845-017-1598-8 pmid: 28508191
[6]	Shilpa, Joshi, Karile P T. A review on exudates detection methods for diabetic retinopathy[J]. Biomedicine & Pharmacotherapy, 2018,9(7):1454-1460.
[7]	Voets M, Mllersen K, Bongo L A. Replication study: Development and validation of deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs[J/OL].(2018-08-30)[2019-07-10] http:∥arxiv.org/abs/1803.04337.
[8]	Wisaeng K, Sa-Ngiamvibool W. Exudates detection using morphology mean shift algorithm in retinal images[J]. IEEE Access, 2019,1(3):11946-11958.
[9]	Parham K, Leandro A, Passos J, et al. Exudate detection in fundus images using deeply-learnable features[J]. Computers in Biology and Medicine, 2019,104(7):62-69.
[10]	Srivastava R, Duan L, Wong D W K, et al. Detecting retinal microaneurysms and hemorrhages with robustness to the presence of blood vessels[J] Computer Methods and Programs in Biomedicine, 2017,138(6):83-91.
[11]	Prenta I P, Lon Ari S. Detection of exudates in fundus photographs using deep neural networks and anatomical landmark detection fusion[J]. Computer Methods and Programs in Biomedicine, 2016,137(3):281-292.
[12]	Feng Z, Yang J, Yao L, et al. Deep retinal image segmentation: a FCN-based architecture with short and long skip connections for retinal image segmentation[C]. Guangzhou:International Conference on Neural Information Processing, 2017.
[13]	Tan J H, Fujita H, Sivaprasad S, et al. Automated segmentation of exudates, haemorrhages, microaneurysms using single convolutional neural network[J]. Information Sciences, 2017,42(6):66-76.
[14]	陈有信, 张碧磊, 张弘哲. 眼科人工智能技术的现状与问题[J]. 中华眼底病杂志, 2019,35(2):119-123.
	Chen Youxin, Zhang Bilei, Zhang Hongzhe. Insights and prospectives of ophthalmologic artificial intelligence technology[J]. Chinese Journal of Ocular Fundus Diseases, 2019,35(2):119-123.
[15]	Zheng R, Liu L. Detection of exudates in fundus photographs with imbalanced learning using conditional generative adversarial network[J]. Biomedical Optics Express, 2018,8(10):56-68.
[16]	Sanchez C I, Garcia M, Mayo A, et al. Retinal image analysis based on mixture models to detect hard exudates[J]. Medical Image Analysis, 2009,13(4):650-658. pmid: 19539518
[17]	段彦华, 周梦颖, 杨春兰, 等. 眼底图像中硬性渗出物检测算法[J]. 北京生物医学工程, 2018,37(1):1-8.
	Duan Yanhua, Zhou Mengying, Yang Chunlan, et al. Algorithm of hard exudates detection in fundus image[J]. Beijing Biomedical Engineering, 2018,37(1):1-8.
[18]	Carla P, Luís G, Manuel F. Exudate segmentation in fundus images using an ant colony optimization approach[J]. Information Sciences, 2015,29(6):14-24.
[19]	João S, Adelino F, Gerardo F. An adaptive hybrid genetic algorithm for pavement management[J]. International Journal of Pavement Engineering, 2019,20(3):266-286.
[20]	徐海, 秦立峰. 遗传算法改进的KSW熵法计算黄瓜叶部角斑病密度[J]. 安徽农业科学, 2017,45(32):212-215.
	Xu Hai, Qing Lifeng. KSW entropy method improved by genetic algorithm for density of cucumber angular leaf spot calculation[J]. Journal of Anhui Agricultural Sciences, 2017,45(32):212-215.
[21]	申小次. 基于遗传算法的二维熵图像分割方法的研究[D]. 北京:中国地质大学, 2008.
	Shen Xiaoci. Two-dimensional entropy image segmentation based on genetic algorithm[D]. Beijing:China University of Geosciences, 2008.
[22]	金元郁, 张洪波, 冯宇. 基于遗传算法的二维双阈值Otsu图像分割算法[J]. 电子科技, 2009,22(11):35-39.
	Jin Yuanyu, Zhang Hongbo, Feng Yu. Two-Dimensional Two-Otsu threshold image segmentation based on the genetic algorithm[J]. Electronic Science and Technology, 2009,22(11):35-39.
[23]	Elbalaoui A, Ouadid Y, Merbouha A. Segmentation of optic disc in fundus images using an active contour[J]. Journal of Electronic Commerce in Organizations, 2018,16(1):97-111. doi: 10.4018/JECO
[24]	姜平. 眼底图像分割方法研究[D]. 长春:吉林大学, 2018.
	Jiang Ping. Study of retinal image segmentation method[D]. Changchun:Jilin University, 2018.
[25]	Saleh M D, Salih N D, Eswaran C, et al. Automated segmentation of optic disc in fundus images[C]. Kuala Lumpur: International Colloquium on Signal Processing & Its Applications, 2014.
[26]	王敏, 童水光, 陈玉辉, 等. 一种基于Hough变换的快速圆检测算法[J]. 机械工程与自动化, 2018(1):152-154.
	Wang Min, Tong Shuiguang, Chen Yuhui, et al. A quick circle detection algorithm based on Hough transformation[J]. Mechanical Engineering & Automation, 2018(1):152-154.
[27]	Diptoneel K, Nju S B. A new dynamic thresholding based technique for detection of hard exudates in digital retinal fundus image[C]. Noida:International Conference on Signal Processing and Integrated Networks(SPIN), 2014.
[28]	Osareh A, Shadgar B, Markham R. A computational-intelligence-based approach for detection of exudates in diabetic retinopathy images[J]. IEEE Transactions on Information Technology in Biomedicine, 2009,13(4):535-545. doi: 10.1109/TITB.2008.2007493 pmid: 19586814
[29]	Xiao Z T, Li F, Geng L, et al. Hard exudates detection method based on background-estimation[C]. Tianjin:International Conference on Image and Graphics, 2015.
[30]	Fraz M M, Jahangir W, Zahid S, et al. Multiscale segmentation of exudates in retinal images using contextual cues and ensemble classification[J]. Biomedical Signal Processing and Control, 2017,35(3):50-62. doi: 10.1016/j.bspc.2017.02.012

方法	Kaggle					DIARETDB1
	基于像素的评价标准		基于图像的评价标准			基于像素的评价标准		基于图像的评价标准
	SE/%	PPV/%	SE/%	SP/%	AC/%	SE/%	PPV/%	SE/%	SP/%	AC/%
文献[29]	—	—	—	—	—	84.6	94.4	97.3	90.0	93.7
文献[30]	—	—	—	—	—	92.4	87.1	—	81.3	87.7
文献[17]	93.2	79.3	97.1	80.0	95.0	—	—	—	—	—
本文算法	83.6	93.2	95.2	86.2	90.8	82.4	93.3	93.6	96.2	89.9