張春劍 戴百生 張恩澤 卜巍 鄔向前
摘要:眼底視網(wǎng)膜血管的分割在眼底視網(wǎng)膜血管病變分析和心腦血管疾病診斷中具有重要的臨床應(yīng)用價(jià)值。針對(duì)現(xiàn)有視網(wǎng)膜血管割算法分割出的血管邊界不夠精確光滑以及對(duì)低對(duì)比度血管分割效果不理想等問題,本文提出一種改進(jìn)的B樣條Ribbon Snake模型,對(duì)視網(wǎng)膜圖像中的血管進(jìn)行分割。該方法首先對(duì)眼底視網(wǎng)膜圖像進(jìn)行亮度均衡化、去噪等預(yù)處理操作,再利用方向線檢測(cè)算子對(duì)血管中心線進(jìn)行提取,最終在傳統(tǒng)B樣條Ribbon Snake模型的基礎(chǔ)上設(shè)計(jì)新的寬度能量、區(qū)域能量,并利用該模型完成對(duì)視網(wǎng)膜血管進(jìn)行分割。實(shí)驗(yàn)結(jié)果表明,該方法分割出的血管邊界具有精確與光滑的特性,且能對(duì)低對(duì)比度血管進(jìn)行有效分割。
關(guān)鍵詞:視網(wǎng)膜; 血管分割; B-樣條; RibbonSnake模型
中圖分類號(hào):TP391 文獻(xiàn)標(biāo)識(shí)碼:A文章編號(hào):2095-2163(2014)04-0032-05
Abstract:The segmentation of retinal blood vessels has an important application value on analysis of retinal vascular lesions and diagnosing cardiovascular diseases. To solve the problem of the poor results of the low-contrast vessels segmentation and rough boundaries for existing retinal blood vessels segmentation algorithms, this paper proposes an improved B-spline Ribbon Snake to segment the retinal vessels. Firstly the paper preprocesses the retinal images through illumination equalization and denoising, and then extracts the vessel centerlines at different directions. Finally, the paper segments retinal blood vessels by designing the width energy and region energy in the B-spline Ribbon Snake model. The results of the experiment shows that the proposed method segmented retinal vessel more accurately and smoothly, and also segments low contrast vessels effectively.
Key words:Retina; Blood Vessels Segmentation; B-spline; RibbonSnake Model
0引言
由于視網(wǎng)膜血管在眼底視網(wǎng)膜血管病變的分析和心腦血管疾病診斷中具有重要意義,多年來一直受到研究人員的高度重視。正常情況下,其結(jié)構(gòu)跟形態(tài)一直處于穩(wěn)定的狀態(tài)。然而高血壓、糖尿病及冠狀動(dòng)脈硬化等嚴(yán)重危害人類身體健康的心腦血管疾病則會(huì)引起眼底視網(wǎng)膜血管直徑和彎曲程度等結(jié)構(gòu)的變化。但是由于視網(wǎng)膜里血管結(jié)構(gòu)復(fù)雜,通過肉眼進(jìn)行檢測(cè)會(huì)導(dǎo)致誤診現(xiàn)象時(shí)有發(fā)生,因此研究科學(xué)有效的分割方法對(duì)眼底視網(wǎng)膜圖像進(jìn)行血管提取即已成為目前研究界的焦點(diǎn)課題之一。
目前視網(wǎng)膜血管分割常用的方法主要有:基于匹配濾波[1]、基于形態(tài)學(xué)[2]以及基于有監(jiān)督學(xué)習(xí)[3-4]的血管分割算法。雖然現(xiàn)有的方法可以較好地分割視網(wǎng)膜血管,但是同樣存在一定的問題而有待改進(jìn),比如分割得到血管邊界不夠精確和低對(duì)比度血管分割的結(jié)果不夠理想。本文則是將B-樣條RibbonSnake模型改進(jìn)之后對(duì)視網(wǎng)膜血管進(jìn)行分割,又通過設(shè)計(jì)適用于視網(wǎng)膜血管分割的寬度能量和區(qū)域能量,達(dá)到分割低對(duì)比度血管目的,并且分割得到的血管邊界光滑、精確。
1視網(wǎng)膜圖像的預(yù)處理
對(duì)于彩色視網(wǎng)膜圖像,其紅色通道上的圖像過飽和,藍(lán)色通道上的圖像對(duì)比度過低,而綠色通道圖像對(duì)比度最好,血管組織同背景差異最為明顯,故本文選擇綠色通道分量作為后續(xù)處理的對(duì)象。
由于眼底視網(wǎng)膜圖像拍攝受到眼底相機(jī)性能和眼球運(yùn)動(dòng)等影響,視網(wǎng)膜圖像不可避免地將存在亮度不均和系統(tǒng)噪聲等問題。針對(duì)圖像亮度不均,本文先對(duì)眼底視網(wǎng)膜圖像采取大尺度中值濾波進(jìn)行過平滑,估計(jì)出圖像的背景亮度,再在原圖基礎(chǔ)上減去圖像背景亮度來實(shí)現(xiàn)亮度的均衡化[5]。考慮到雙邊濾波器[6-7]在去除圖像噪聲的同時(shí),還能更進(jìn)一步地保護(hù)血管邊緣和細(xì)節(jié),因此研究采用雙邊濾波器對(duì)亮度均衡化后的血管圖像進(jìn)行去噪。預(yù)處理結(jié)果如圖1所示。
2視網(wǎng)膜血管中心線的提取
接下來,本文將在預(yù)處理之后的圖像上,實(shí)現(xiàn)對(duì)視網(wǎng)膜血管中心線的提取??紤]到視網(wǎng)膜血管是一種屋脊邊緣,其灰度截面輪廓呈高斯分布,研究將利用方向線檢測(cè)算子[8]來提取視網(wǎng)膜中心線。本文在不同的方向?qū)σ暰W(wǎng)膜血管中心線進(jìn)行檢測(cè),并將得到的結(jié)果融合形成完整的視網(wǎng)膜血管中心線。在此,可將在方向θ上檢測(cè)到的中心線稱為θ-方向線,而將檢測(cè)θ-方向線的檢測(cè)算子稱為θ-方向線檢測(cè)算子。
5結(jié)束語(yǔ)
本文對(duì)眼底視網(wǎng)膜圖像進(jìn)行亮度均衡化、去噪等預(yù)處理之后,在各個(gè)方向上對(duì)血管中心線進(jìn)行有效提取,并通過設(shè)計(jì)寬度能量和區(qū)域能量,對(duì)B-樣條RibbonSnake模型進(jìn)行改進(jìn),最終完成對(duì)視網(wǎng)膜血管的精確分割。本文在DRIVE數(shù)據(jù)庫(kù)進(jìn)行血管分割測(cè)試,結(jié)果表明本文方法分割出的血管邊界精確、光滑,且能對(duì)低對(duì)比度血管實(shí)現(xiàn)有效分割。雖然本文方法能夠精確分割出眼底視網(wǎng)膜圖像中絕大多數(shù)血管,但是對(duì)細(xì)小血管的分割效果仍不理想,如何對(duì)細(xì)小血管實(shí)現(xiàn)完整分割也就隨之成為下一步的重點(diǎn)研究方向。
參考文獻(xiàn):
[1]CHAUDHURI S,CHATTERJEE S, KATZ N, et al. Detection of blood vessels in retinal images using two-dimensional matched filters[J]. IEEE Trans. Med. Imag., 1989, 8(3):263-269.
[2]MENDONCA A M, CAMPILBO A. Segmentation of retinal blood vessels by combining the detection of centerlines and morphological reconstruction[J]. IEEE Trans. Med. Imag., 2006, 25(9):1200-1213.(下轉(zhuǎn)第38頁(yè))
[3]RICCI E, PERFETTI R. Retinal blood vessel segmentation using line operators and support vector classification[J]. IEEE Trans. Med. Imag., 2007, 26(10):1357-1365.
[4]SOARES J V B, LEANDRO J J G, CEASR R M, et al., Retinal vessel segmentation using the 2-D Gabor wavelet and supervised classification[J]. IEEE Trans. Med. Imag., 2006, 25(9):1214-1222.
[5]NIEMEIJER M, et al. Automatic detection of red lesions in digital color fundus photographs[J]. Medical Imaging, IEEE Transactions on, 2005, 24(5): 584-592.
[6]LIN C, TSAI J, CHIU C. Switching bilateral filter with a texture/noise detector for universal noise removal[J]. Image Processing, IEEE Transactions on, 2010, 19(9): 2307-2320.
[7]邱宇.基于雙邊濾波的圖像去噪及銳化技術(shù)研究[D]. 重慶:重慶大學(xué), 2011.
[8]鄔向前,等. 掌紋紋線提取及病理紋的識(shí)別[C]//計(jì)算機(jī)在診法中的應(yīng)用于研究學(xué)術(shù)研討會(huì)論文匯編. 2005:99-104.
[9]KASS M, WITKIN A, TERZOPOULOS D. Snakes: Active contour models[J]. International Journal of Computer Vision, 1988, 1(4): 321-331.
[10]LAPTEV I, MAYER H, LINDEBERG T, et al. Automatic extraction of roads from aerial images based on scale space and snakes[J].Machine Vision and Applications, 2000,12(1): 23-31.
[11]胡陽(yáng), 等. 一種改進(jìn)的Ribbon Snake遙感圖像道路自動(dòng)生成算法[J].計(jì)算機(jī)應(yīng)用, 2009, 29(3):747-749.
[12]SCHMUGGE, STEPHEN J, et al. Segmentation of vasculature for intravital microscopy using bridging vessel snake[C]//Biomedical Imaging, 3rd IEEE International Symposium on. IEEE, 2006:177-180.
[13]STAALJ J, ABRAMOFF M D, NIEMEIJER M, et al. Ridge based vessel segmentation in color images of the retina[J], IEEE Trans. Med.Imag., 2004, 23:501-509.
[14]MARTINEZ-PeREZ M E, et al. Retinal blood vessel segmentation by means of scale-space analysis and region growing[C].Medical Image Computing and Computer-Assisted Intervention-MICCAI, 1999:90-97.