基于免疫遺傳形態學的視網膜光學相干斷層圖像邊緣

童何俊; 付冬梅

doi:10.13374/j.issn2095-9389.2019.04.015

基于免疫遺傳形態學的視網膜光學相干斷層圖像邊緣

doi: 10.13374/j.issn2095-9389.2019.04.015

童何俊,
付冬梅^,

北京科技大學自動化學院，北京 100083

詳細信息

通訊作者:
付冬梅, E-mail: fdm_ustb@ustb.edu.cn

中圖分類號: TP391.41
計量
- 文章訪問數: 904
- HTML全文瀏覽量: 405
- PDF下載量: 13
- 被引次數: 0
出版歷程
- 收稿日期: 2018-03-01
- 刊出日期: 2019-04-15

Edge detection method of retinal optical coherence tomography images based on immune genetic morphology

TONG He-jun,
FU Dong-mei^,

School of Automation and Electrical Engineering, University of Science and Technology Beijing, Beijing 100083, China

More Information

Corresponding author: FU Dong-mei, E-mail: fdm_ustb@ustb.edu.cn

摘要

摘要: 提出了基于免疫遺傳算法的形態學自適應結構元素生成算法，并將其用于光學相干斷層成像(optical coherence tomography，OCT) 圖像中視網膜組織邊緣檢測. 首先將圖像進行去噪和粗分割的預處理，并將圖像劃分為若干子圖像; 其次對每一子圖利用免疫遺傳算法求取自適應結構元，初始隨機生成固定長度的二進制數串作為抗體，并將其轉化為結構元素格式，以圖像二維熵定義抗體適應度，根據子圖像本身結構特征信息，尋找最優抗體結構元素; 最后利用尋優得到的各結構元素對子圖進行形態學邊緣檢測，合并各子圖的分割結果，實現整體圖像目標邊界提取. 實驗結果表明了該方法在圖像目標邊界提取的有效性.
- 數學形態學 /
- 自適應結構元素 /
- 免疫遺傳算法 /
- 光學相干斷層圖像 /
- 邊緣檢測
Abstract: Optical coherence tomography (OCT) is an indispensable tool used for the diagnosis and identification of ocular fundus disease and nondestructive, rapid, and high-resolution imaging of the living retinas. The attendant research focuses on the development of computer-aided methods to help ophthalmologists make judgments regarding the morphological changes of retinal tissue and acquire tissue characteristic parameters. Realizing the segmentation of retinal tissue in OCT images is the key aspect of this kind of research. Mathematical morphology, which has been widely used in the fields of image detection, shape analysis, pattern recognition, and computer vision, uses different structural elements to measure, extract, analyze, and identify image targets. However, traditional morphological structure elements cannot be adaptively changed on the basis of the structural characteristics of the images. In this study, an algorithm for generating morphological adaptive structural elements was proposed on the basis of an immune genetic algorithm, which the detection of retinal tissue edges in optical coherence tomography (OCT) images was applied. First, the image is preprocessed by denoising and coarse segmentation and then the image is divided into several sub-images. Second, the adaptive structure elements are computed using an immune genetic algorithm for each sub-image. A string of binary numbers of fixed length is initially randomly generated as an antibody and then converted into a format of structural element. The fitness of an antibody is defined by the two-dimensional entropy of the image and the optimal antibody and structural elements are identified according to the structural characteristics of the subimage itself. Finally, with these optimal structural elements, morphological edge detection is performed to obtain the segmentation results of each sub-image combined with those of each sub-graph to realize the extraction of the target boundary of the whole image. The experimental results show the proposed method to be effective in the boundary extraction of images.
- mathematical morphology /
- adaptive structural elements /
- immune genetic algorithm /
- optical coherence tomography /
- edge detection

HTML全文

圖 1 免疫遺傳算法程序框圖

Figure 1. Flowchart of immune genetic algorithm

下載: 全尺寸圖片幻燈片

圖 2 鏈式抗體與結構元素之間的轉化關系

Figure 2. Transformation between chain antibodies and structural elements

下載: 全尺寸圖片幻燈片

圖 3 視網膜邊界提取過程. (a) 原圖; (b) 去噪預處理; (c) 視網膜區域粗分割; (d) 圖像分塊; (e) 自適應結構元與邊緣檢測; (f) 最終分割結果

Figure 3. Extraction process of retinal boundary: (a) original image; (b) de-noising preprocessing; (c) rough segmentation of retinal region; (d) image segmentation into blocks; (e) adaptive structure element and edge detection; (f) final segmentation result

下載: 全尺寸圖片幻燈片

圖 4 多種邊界提取方法結果比較. (a) 原圖; (b) 去噪預處理; (c) 專家手動分割結果; (d) 尺度M = 3結構元; (e) 尺度M = 5結構元; (f) 多尺度結構元; (g) Canny; (h) GTDP; (i) IGM; (j) IGSM

Figure 4. Comparison of multiple boundary extraction methods: (a) original image; (b) de-noising preprocessing; (c) expert manual segmentation results; (d) structure element scale M = 3; (e) structure element scale M = 5; (f) multiscale structural element; (g) Canny; (h) GTDP; (i) IGM; (j) IGSM

下載: 全尺寸圖片幻燈片

圖 5 黃斑中心凹區域劃分示意圖

Figure 5. Schematic map of the division of the macular fovea

下載: 全尺寸圖片幻燈片

表 1 多種邊界提取方法品質因數評價結果

Table 1. Partt quality factor evaluation results by multiple boundary extraction methods

邊界提取方法	品質因數，R/%
Single_3	88. 31
Single_5	75. 79
Multi	84. 71
Canny	43. 05
GTDP	84. 7
IGM_3	89. 85
IGM_5	85. 9
IGSM_3	90. 23
IGSM_5	87. 23

下載: 導出CSV

表 2 視網膜組織量化結果

Table 2. Quantitative results of retinal tissue

邊界提取方法	區域	A/S1	A/S2	A/S3	A/S4	A/S5	T/S1	T/S2	T/S3	T/S4	T/S5
專家手動分割	NSL_R	1524	928	90	1051	1680	26.74	14.28	2.31	16.17	29.47
	NSL_G	932	1309	878	1256	893	16.35	20.14	22.51	19.32	15.67
	RPE	600	730	468	768	614	10.53	11.23	12	11.82	10.77
Single_3	NSL_R	1502	923	106	1055	1652	26.35	14.20	2.72	16.23	28.98
	NSL_G	948	1320	861	1254	912	16.63	20.31	22.08	19.29	16.00
	RPE	454	714	468	759	606	7.96	10.98	12.00	11.68	10.63
Single_5	NSL_R	1377	777	43	1012	1529	24.16	11.95	1.10	15.57	26.82
	NSL_G	1069	1452	922	1823	1029	18.75	22.31	23.64	19.74	18.05
	RPE	459	579	388	637	485	8.05	8.91	9.95	9.80	8.51
Multi	NSL_R	1506	930	117	1056	1652	26.42	14.31	3.00	16.25	28.98
	NSL_G	944	1302	852	1252	910	16.56	20.03	21.85	19.26	15.96
	RPE	587	724	468	761	607	10.30	11.14	12.00	11.71	10.65
Canny	NSL_R	1522	950	162	1085	1669	26.7	14.62	4.15	16.69	29.28
	NSL_G	913	1275	787	1218	884	16.02	19.62	20.18	18.74	15.51
	RPE	609	729	492	768	617	10.68	11.22	12.62	11.82	10.82
GTDP	NSL_R	1151	1098	814	1240	1659	26.51	16.89	20.87	19.08	29.11
	NSL_G	930	1131	156	1063	896	16.32	17.40	4.00	16.35	15.72
	RPE	595	725	468	761	619	10.44	11.15	12.00	11.71	10.86
IGM	NSL_R	1501	921	105	1059	1652	26.33	14.17	2.69	16.29	28.98
	NSL_G	948	1323	862	1251	911	16.63	20.35	22.10	19.25	15.98
	RPE	582	713	468	759	606	10.21	10.97	12.00	11.68	10.63
IGSM	NSL_R	1501	923	107	1059	1651	26.33	14.20	2.74	16.29	28.96
	NSL_G	949	1322	860	1251	913	16.65	20.34	22.05	19.25	16.02
	RPE	584	713	468	759	606	10.25	10.97	12.00	11.68	10.63

下載: 導出CSV

表 3 多種邊界提取方法與專家結果量化平均相對誤差比較(以面積為例)

Table 3. Comparison of the average relative errors determined by multiple boundary extraction methods and expert results (as an example of area)

邊界提取方法	Single_3	Single_5	Multi	Canny	GTDP	IGM	IGSM
ARE	0.0383	0.1586	0.0303	0.0751	0.6385	0.0241	0.0256

下載: 導出CSV

259luxu-164

參考文獻(16)

[1]	Haralick R M, Sternberg S R, Zhuang X H. Image analysis using mathematical morphology. IEEE Trans Pattern Anal Mach Intell, 1987, 9(4): 532 http://europepmc.org/abstract/MED/21869411
[2]	Liu Y L, Gui Z G. Adaptive image enhancement algorithm with variable weighted matching based on morphology. J Electron Inform Technol, 2014, 36(6): 1285 https://www.cnki.com.cn/Article/CJFDTOTAL-DZYX201406003.htm 劉艷莉, 桂志國. 基于形態學的可變權值匹配自適應圖像增強算法. 電子與信息學報, 2014, 36(6): 1285 https://www.cnki.com.cn/Article/CJFDTOTAL-DZYX201406003.htm
[3]	Zhou S B, Shen A Q, Li G F. Concrete image segmentation based on multiscale mathematic morphology operators and Otsu method. Adv Mater Sci Eng, 2015, 2015(11): 208473
[4]	Lee S H, Lee C. Multiscale morphology based illumination normalization with enhanced local textures for face recognition. Expert Syst Appl, 2016, 62: 347 doi: 10.1016/j.eswa.2016.06.039
[5]	Wang J G, Gao D Y.Improved morphological TOP-HAT filter optimized with genetic algorithm//Proceedings of the 2009 2nd International Congress on Image and Signal Processing (CISP).Tianjin, 2009: 1
[6]	Babu K R, Sunitha K V N. Image de-noising and enhancement for salt and pepper noise using genetic algorithm-morphological operations. Int J Signal Image Process, 2013, 4(1): 36 http://www.researchgate.net/publication/236166772_Image_De-noising_and_Enhancement_for_Salt_and_Pepper_Noise_using_Genetic_Algorithm-Morphological_Operations
[7]	Erc?al T, ?zcan E, Asta S.Soft morphological filter optimization using a genetic algorithm for noise elimination//Proceedings of 2014 14th UK Workshop on Computational Intelligence.Bradford, 2014: 1
[8]	Jiang D H, Hua G. Research on image enhancement method based on adaptive immune genetic algorithm. J Comput Theor Nanosci, 2015, 12(1): 119 doi: 10.1166/jctn.2015.3707
[9]	Chen L A, Zhang P M. Realization of immune genetic algorithm in MATLAB. J Fuzhou Univ Nat Sci, 2004, 32(5): 554 doi: 10.3969/j.issn.1000-2243.2004.05.010 陳麗安, 張培銘. 免疫遺傳算法在MATLAB環境中的實現. 福州大學學報(自然科學版), 2004, 32(5): 554 doi: 10.3969/j.issn.1000-2243.2004.05.010
[10]	Liu Y N, Zhang G C, An J. Noise-resistance in color image edge detection based on flexible morphology. J Lanzhou Univ Nat Sci, 2016, 52(1): 135 https://www.cnki.com.cn/Article/CJFDTOTAL-LDZK201601023.htm 劉燕妮, 張貴倉, 安靜. 基于柔性形態學的抗噪彩色圖像邊緣檢測. 蘭州大學學報(自然科學版), 2016, 52 (1): 135 https://www.cnki.com.cn/Article/CJFDTOTAL-LDZK201601023.htm
[11]	Bhima K, Jagan A.Analysis of MRI based brain tumor identification using segmentation technique//Proceedings of 2016 International Conference on Communication and Signal Processing.Melmaruvathur, 2016: 2109
[12]	Michikawa T, Suzuki H, Moriguchi M, et al. Automatic extraction of endocranial surfaces from CT images of crania. PloS One, 2017, 12(4): e0168516 http://europepmc.org/abstract/MED/28406901
[13]	Xu D L, Tan R Q, Ding Q, et al. Screening ocular fundus macular diseases with OCT. Clin J Chin Med, 2016, 8(32): 123 https://www.cnki.com.cn/Article/CJFDTOTAL-ZYLY201632064.htm 徐黛麗, 譚榮強, 丁瓊, 等. 眼底黃斑部疾病篩查中OCT的應用. 中醫臨床研究, 2016, 8(32): 123 https://www.cnki.com.cn/Article/CJFDTOTAL-ZYLY201632064.htm
[14]	Li Y J, Zhang J W, Wang M N. Improved BM3D denoising method. IET Image Process, 2017, 11(12): 1197 doi: 10.1049/iet-ipr.2016.1110
[15]	Chiu S J, Allingham M J, Mettu P S, et al. Kernel regression based segmentation of optical coherence tomography images with diabetic macular edema. Biomed Opt Exp, 2015, 6(4): 1172 doi: 10.1364/BOE.6.001172
[16]	Fu D M, Tong H J, Zheng S, et al. Retinal status analysis method based on feature extraction and quantitative grading in OCT images. Biomed Eng Online, 2016, 15(1): 87 doi: 10.1186/s12938-016-0206-x