北京生物醫(yī)學(xué)工程

用于三維醫(yī)學(xué)圖像配準(zhǔn)的“粗精”混合算法研究_

A ‘coarse and fine’ hybrid algorithm for three-dimensional medical image registration

作者：翁飛侯文廣曾明平

單位：武漢大學(xué)中南醫(yī)院設(shè)備處(武漢430071) 

關(guān)鍵詞：圖像配準(zhǔn)；主成分分析；Powell優(yōu)化算法

分類號：R318.04;TP391.9

出版年·卷·期（頁碼）：2016·35·1（12-17）

摘要：

目的三維醫(yī)學(xué)圖像配準(zhǔn)能夠?yàn)榕R床診斷提供更多更豐富的信息，是醫(yī)學(xué)圖像處理領(lǐng)域的研究熱點(diǎn)。本文針對配準(zhǔn)中傳統(tǒng)方法很難兼顧到配準(zhǔn)的準(zhǔn)確度和速度，提出一種基于“粗精”混合配準(zhǔn)的算法。方法首先，采用主成分分析方法對圖像進(jìn)行粗配準(zhǔn)，減小浮動圖像和參考圖像之間的差異，得到精配準(zhǔn)良好的初始參數(shù)；然后，采用改進(jìn)的Powell算法在初始參數(shù)的基礎(chǔ)上進(jìn)行精配準(zhǔn)；最后，以三維MRI圖像為例設(shè)計(jì)了兩組實(shí)驗(yàn)進(jìn)行驗(yàn)證。結(jié)果該方法配準(zhǔn)精度高，旋轉(zhuǎn)參數(shù)誤差低至0.001，得到的圖像灰度誤差可限制在2以內(nèi)。此外與全局優(yōu)化方法相比，該方法保證配準(zhǔn)精度的同時(shí)，在速度上可提高2～3倍。結(jié)論實(shí)驗(yàn)證明該方法可同時(shí)兼顧配準(zhǔn)的準(zhǔn)確度和速度。

Objective Three-dimensional medical image registration can provide more comprehensive information for clinical diagnosis and is a hot field of medical image processing. Traditional registration methods are difficult to take into account the accuracy and speed of registration at the same time. To solve the problem,this paper proposes an method based on ‘coarse and fine’ hybrid registration algorithm. Methods First,principal component analysis was used for coarse image registration in order to reduce the difference between floating images and the reference images and get good initial fine alignment parameters. Then,improved Powell algorithm was used for refined registration based on initial parameters. Finally,two groups of experiments based on 3D MRI images were designed to verify the method. Results The registration accuracy of this method was high,in which the rotation parameter error was about 0.001 while the image gray level error was limited to 2. Additionally,this method could be 2 to 3 times faster than the global method while the registration accuracy not be reduced. Conclusions Experiments illustrated the registration accuracy and speed could be taken into account at the same time in this method.

【Keywords】

參考文獻(xiàn)：

［1］俞亞青,田學(xué)隆,閆春紅. 醫(yī)學(xué)圖像配準(zhǔn)方法分類及現(xiàn)狀［J］. 重慶大學(xué)學(xué)報(bào):自然科學(xué)版,2003,26(8):114-118.

Yu Yaqing,Tian Xuelong,Yan Chunhong. Classification and state of medical image registration methods［J］. Journal of Chongqing University:Natural Science Edition,2003,26(8):114-118.

［2］羅述謙,周果宏. 醫(yī)學(xué)圖像處理與分析［M］. 北京:科學(xué)出版社,1992：140-141.

［3］van den Elsen PA,pol EJD,Viergever MA,et al. Medical image matching a review with classification ［J］. IEEE Engineering in Medical and Biology,1993,12(4):26-39.

［4］Maurer CR,Fitzpatrick J. A review of medical image registration［M］//Mociunas ed. Park ridge,IL,AANS:Interactive Image Guided Neurosurgery,1993:17-44.

［5］郭科，陳聆，魏友華.最優(yōu)化方法及其應(yīng)用［M］. 北京:高等教育出版社,2007.

［6］Jolliffe I. Principal Component Analysis［M］. Hoboken:New Jersey:John Wiley & Sons,Ltd,2002.

［7］李了了,鄧善熙,丁興號. 基于大津法的圖像分塊二值化算法［J］. 微計(jì)算機(jī)信息,2005,21(8-3):76-77.

Li Liaoliao,Deng Shanxi,Ding Xinghao. Binarization algorithm based on image partition derived from Da-Jing method ［J］. Control & Automation,2005,21(8-3):76-77.

［8］Powell MJD. On the convergence of the variable metric algorithm［J］. IMA Journal of Applied Mathematics,1971,7(1):21-36.

［9］何利民. Powell 方法的收斂速度［J］. 復(fù)旦學(xué)報(bào)：自然科學(xué)版,1988,(3):68-75.

He Limin. Convergence rate of Powell method［J］. Journal of Fudan University:Natural Science,1988,(3):68-75.

［10］Fogel GB,Corne DW.Evolutionary Computation in Bioinformatics［M］. San Francisco:Morgan Kaufmann,2002.

［11］郭變芳. 三維醫(yī)學(xué)圖像剛體配準(zhǔn)［D］. 武漢：華中科技大學(xué),2009.

Guo Bianfang. 3D rigid medical image registration［D］. Wuhan:Huazhong University of Science & Technology,2009.

服務(wù)與反饋：

【文章下載】【加入收藏】

提示：您還未登錄，請登錄！點(diǎn)此登錄

51黑料吃瓜在线观看,51黑料官网|51黑料捷克街头搭讪_51黑料入口最新视频