基于模糊聚类的尿沉渣有形成分分析研究
|
摘要
尿沉渣图像检验作为临床上病理分析的重要依据,已经成为医学研究讨论的重要话题。尿有形成分检查的重要性在于它是尿液分析中不可缺少的检查手段,对临床诊断、治疗监测及健康普查具有重要的临床意义,对肾脏疾病、泌尿道疾病、循环系统疾病以及感染性疾病等,有重要的诊断和鉴别作用,尿沉渣图像检验以及其有形成分分析的准确性和分析速度的快慢就成为尿沉渣图像研究的焦点问题。很显然,以往的人工肉眼检验尿沉渣图像有形成分的准确率的确比较高。可是各大医院做尿检的份额原本就惊人,更可怕的是尿沉渣图像中成分繁多,个体目标也是复杂多变,人工分析难以满足要求。借助图像处理技术实现自动分析是解决这一问题的好办法。
全自动尿沉渣分析仪是一种高智能、全自动、客观的基于计算机显微图像的尿液有形成份分析仪器。它集计算机技术、精密机械技术、光学显微成像技术、自动控制技术、数字图像处理与机器视觉技术于一体。而这其中,数字图像的处理和理解是核心技术之一。尿沉渣有形成分的模式识别方法有多种,如人工神经网络识别,支持向量机识别,贝叶斯方法等等。但是这些方法无一例外采用是的线性的识别方法。
本文所述的方法中没有要求对每个尿沉渣个体进行识别,而是首先提取了目标的5个形态特征和12个纹理特征。以这些特征参数为依据,用模糊聚类的方法将所有的目标聚成几类。聚类分析是一种典型的数据挖掘和分析的方法,其中关于聚类类别的最终确定采用了F-统计量法。聚类之后的所有类不可能是没有杂类元素的纯类,用类间阈值分割(Ostu)法可以去除一些类中与大多数个体相似度较低的元素,并且把它们加入到待定集合当中。这样做的目的一是保证后面给每类定性时的准确率,二是将这些不定元素可以进行重新识别以提高准确率。然后将每类中被抽取出来的元素经过神经网络的检验,由它们代表整个类来给类定性是红细胞、白细胞、管型、还是其它结晶。最后对处于待定集合中的个体进行重新识别,以提高识别准确率。
经过数据实验和仿真证明了该方法的可行性和有效性。对于大小面积差异明显的个体准确率较高,如管型和上皮细胞。草酸钙结品的纹理特征明显,该法也有一定优势。对于差异较小的红细胞和白细胞则效果一般。
Urine sediment image checking as the important evidence clinical pathologic analysis has become a key topic of medical research discussion. Urine visual component checking is very essential because it is incredible checking tool in urine analysis which is not only clinically significant to clinic diagnose, remedial monitoring and health investigation, but also playing an important role in diagnose and differentiation of kidney diseases, urinary tract diseases, circulatory system diseases and infectious diseases. Therefore, the veracity and speed of urine sediment image checking and visual component analysis is the key point to urine sediment image research. Obviously, traditional manual naked eye checking definitely has higher veracity. Despite the huge mount of urine specimens in every large hospital, multiple types are in urine sediment image, what is worse, each of objectives are also complicated. Such huge problem can be settled by urine sediment analyzer and digital image processing.
Nowadays, automatic urine sediment analyzer is a high intelligent, automatic, objectively based on PC micro image urine visual component analyzer. It is a precise apparatus system combining PC technology, precise mechanical technology, and optical micro imaging technology, auto control technology, digital image processing technology and visual machine technology. Among all these above, digital image processing and understanding is one of the key points. There are multiple methods in urine sediment visual components model identification, such as ANN, SVM, Bayers and etc. But none of these is exceptionally adopting the pattern of linear recognizing method.
The method abandoned reorganizing every objective is first pick 5 shape traits and 12 texture traits of objectives. Depending on these trait parameters, gather all objectives into several sorts using fuzzy clustering. Clustering Analysis is a typical data explore and analysis method. Use F-statistic to determine the final clustering. After clustering, use Otsu to take out the elements with low similarity and take them into indeterminate set, because impurity in each sort is inevitable. Thus, we guarantee the veracity determination correctness and raise the veracity by re-identifying the uncertain elements. Then, put the sampling elements from each sorts through the ANN to give the decided component name to each sorts. The elements will decide which sort is red cell and which white cell is. At last, use ANN to re-identify those uncertain elements to raise veracity.
The method is proved valid and effective by experiments and data. As to the elements with greater discrepancy in area veracity is high, such as casts and epithelial cells. It is also easy to distinguish calcium oxalate crystals, but ordinary effort in smaller discrepancy cells.
全自动尿沉渣分析仪是一种高智能、全自动、客观的基于计算机显微图像的尿液有形成份分析仪器。它集计算机技术、精密机械技术、光学显微成像技术、自动控制技术、数字图像处理与机器视觉技术于一体。而这其中,数字图像的处理和理解是核心技术之一。尿沉渣有形成分的模式识别方法有多种,如人工神经网络识别,支持向量机识别,贝叶斯方法等等。但是这些方法无一例外采用是的线性的识别方法。
本文所述的方法中没有要求对每个尿沉渣个体进行识别,而是首先提取了目标的5个形态特征和12个纹理特征。以这些特征参数为依据,用模糊聚类的方法将所有的目标聚成几类。聚类分析是一种典型的数据挖掘和分析的方法,其中关于聚类类别的最终确定采用了F-统计量法。聚类之后的所有类不可能是没有杂类元素的纯类,用类间阈值分割(Ostu)法可以去除一些类中与大多数个体相似度较低的元素,并且把它们加入到待定集合当中。这样做的目的一是保证后面给每类定性时的准确率,二是将这些不定元素可以进行重新识别以提高准确率。然后将每类中被抽取出来的元素经过神经网络的检验,由它们代表整个类来给类定性是红细胞、白细胞、管型、还是其它结晶。最后对处于待定集合中的个体进行重新识别,以提高识别准确率。
经过数据实验和仿真证明了该方法的可行性和有效性。对于大小面积差异明显的个体准确率较高,如管型和上皮细胞。草酸钙结品的纹理特征明显,该法也有一定优势。对于差异较小的红细胞和白细胞则效果一般。
Urine sediment image checking as the important evidence clinical pathologic analysis has become a key topic of medical research discussion. Urine visual component checking is very essential because it is incredible checking tool in urine analysis which is not only clinically significant to clinic diagnose, remedial monitoring and health investigation, but also playing an important role in diagnose and differentiation of kidney diseases, urinary tract diseases, circulatory system diseases and infectious diseases. Therefore, the veracity and speed of urine sediment image checking and visual component analysis is the key point to urine sediment image research. Obviously, traditional manual naked eye checking definitely has higher veracity. Despite the huge mount of urine specimens in every large hospital, multiple types are in urine sediment image, what is worse, each of objectives are also complicated. Such huge problem can be settled by urine sediment analyzer and digital image processing.
Nowadays, automatic urine sediment analyzer is a high intelligent, automatic, objectively based on PC micro image urine visual component analyzer. It is a precise apparatus system combining PC technology, precise mechanical technology, and optical micro imaging technology, auto control technology, digital image processing technology and visual machine technology. Among all these above, digital image processing and understanding is one of the key points. There are multiple methods in urine sediment visual components model identification, such as ANN, SVM, Bayers and etc. But none of these is exceptionally adopting the pattern of linear recognizing method.
The method abandoned reorganizing every objective is first pick 5 shape traits and 12 texture traits of objectives. Depending on these trait parameters, gather all objectives into several sorts using fuzzy clustering. Clustering Analysis is a typical data explore and analysis method. Use F-statistic to determine the final clustering. After clustering, use Otsu to take out the elements with low similarity and take them into indeterminate set, because impurity in each sort is inevitable. Thus, we guarantee the veracity determination correctness and raise the veracity by re-identifying the uncertain elements. Then, put the sampling elements from each sorts through the ANN to give the decided component name to each sorts. The elements will decide which sort is red cell and which white cell is. At last, use ANN to re-identify those uncertain elements to raise veracity.
The method is proved valid and effective by experiments and data. As to the elements with greater discrepancy in area veracity is high, such as casts and epithelial cells. It is also easy to distinguish calcium oxalate crystals, but ordinary effort in smaller discrepancy cells.
引文
[1]柯行斌,王汝传.白细胞图像分割的研究与实现[J].南京邮电学院学报,2003.9
[2]张勇,张强,虞烈.真彩色血细胞显微图像自动识别系统研究[J].西安交通大学学报,1999.33(2)
[3]刘圆圆.医学显微图像工作站[M].浙江:浙江大学通信与信息系统,2004
[4]吕小龙.尿沉渣有形成分自动分类识别算法研究[D].南京,东南大学生物医学工程系,2006.5
[5]李庆,韩红梅,耿鑫金.全自动尿沉渣检验误诊的综合因素[J].中国误诊学杂志,2002.10(2)
[6]王昌富,丛玉隆.尿液有形成分流式细胞分析技术及应用策略[J].国外医学临床生物化学与检验学分册,2003.24(2)
[7]陈红兵.尿液分析仪检测假阴性与假阳性原因探讨[J]浙江临床医学,2002.12(4)
[8]蒋红兵等.计算机显微图像尿沉渣分析仪的研究[J].医疗卫生装备,2001.4
[9]王选贺.基于尿沉渣显微图像的模式识别[D].长春,吉林大学,2007.4
[10]杜莉莉.基于分水岭算法的尿沉渣有形成分自动分类识别研究[D].长春,吉林大学,2007.4
[11]顾可梁.尿有形成分的识别与检查方法的选择[J].中华检验医学杂志,2005.28(6)
[12]Schanze T.Sinc Interpolation of discrete periodic signals.IEEE Transactions on Signal Processing,1999.47(11)
[13]马骏龙,丛玉隆,陈淑云等.全自动尿沉渣分析仪检测尿红细胞鉴别血尿来源的探讨[J].中华医学检验杂志,1997.20(5)
[14]顾可梁.尿沉渣镜检[J].上海医学检验杂志,2000.15(2)
[15]李思改,李卫,江志兰.尿液有形成分检查.尿化学分析法,尿人工显微镜检查UF-100尿沉渣全自动分析仪的比较[J].广东山市第二人民医院检验科,2001.12(2)
[16]尿显微镜:它的临床价值,澳洲悉尼皇室北海滨医院大学,内科、肾科,Sysmes Journal International Vol.6 No.1 1996
[17]谢华.基于多分类器融合的骨髓细胞识别技术研究[J].浙江大学硕士生毕业论文,2005
[18]蔡永军,刘伟玲,虞启琏.遗传神经网络在尿沉渣识别中的应用[J].医疗卫生装备,2004.1
[19]Beni C,Liu X M.A least biased fuzzy clustering method.IEEE Trans.PAMI1992,16(9)
[20]苗雪兰,刘瑞新,宋歌.计算机图形学理论及应用技术[M].北京:机械工业出版社,2007.2
[21]Zhang D S,Lu G J.Segmentation of moving objects in image sequences:Intelligence,2003.12(5)
[22]Kayargadde V,Martens J B.Estimation of edge parameters and image blur using polynomial transforms.CVGIP-GMIP,1996.56(6)
[23]Castleman K R.Digital Image Processing:Prentice-Hall,1996
[24]Cross G,Jain A.Markov random field texture models.IEEE Transactions on Pattern Analysis and Machine Intelligence,1983.5(1)
[25]Milan Sonka,Vaclav Hlavac,Roger Boyle.图像处理、分析与机器视觉[M].北京:人民邮电出版社,2003
[26]赵峰,赵荣椿.纹理分割及特征提取方法综述[J].中国体视学与图像分析,1998.3(4)
[27]钟珞等.模式识别[M].武汉:武汉大学出版社,2006
[28]基于神经网络的尿沉渣有形成分自动分类识别研究[D].杭州:浙江大学,2006.5
[29]章毓晋.图像工程(中)图像分析[M].北京:清华大学出版社,2005.10
[30]Zadeh.L A.Fuzzy sets[J].Information and Control,1965.8
[31]Zadeh.LA.Outline of a new approach to the analysis of complex systems and decision processes,IEEE Trans.Syst.,Man,Cybern,vol.SMC-3,1973
[32]刘普寅,吴孟达,模糊理论及其应用,长沙:国防科技大学出版社,2001.12
[33]何新贵,模糊知识处理的理论与技术,北京:国防工业出版社,1999.7
[34]L.A.Zadeh著,模糊集合-语言变量及模糊逻辑,北京:科学出版社,1982.5
[35]Ostu N.A,Threshold Selection Method from Gray-level Histogram,IEEE trans System Man Cybernetic,1978(8)
[36]夏良正.数字图象处理[M].南京:东南大学出版社,1999
[37]飞思科技产品研发中心,.Visual C++.NET编程指南[M].北京:电子工业出版社.2003,2-10
[38]K.R.Castleman著.朱志刚,石定机等译.数字图像处理.电子工业版,1998
[39]S.Arivazhagan,L.Ganesan.Texture segmentation using wavelet transform.Pattern recognition Letters,2003,24(23)
[40]Vincent Barra.Robust segmentation and analysis of DNA microarray spots using an adaptative split and merge algorithm.Computer Methods and Programs in Biomedicine,2006,81(2)
[41]边肇祺.张学工等,模式识别[M].北京:清华大学出版社.2000
[42]周长发.精通VisualC++图像编程[M].北京:电子工业出版社.2000
[43]Rafael C.Gonzalez Richard E.Woods.Digital Image Processing Second Edition[M].Beijing:Publishing House of Electronics Industry,2002.7
[44]W.Smith,The Scientist and Engineer's and Guide to Digital Signal Processing,California Technical Publishing,1997
[45]Haralick R M,Shapiro L G.Image segmentation techniques.Computer Vision,Graphics and Image Processing,1985
[46]Kittler J,Illingworth J,Foglein J.Threshold selection based on a simple image statistics.Computer Vision,Graphics and Image Processing,1985:30
[47]W.Ying,Y.Nakano,O.Nagata"A Method of Classifying Blood Cells in Urinary Sediments using Neural Network",Medical Imaging Technology,1998:2
[48]Ning Feng ZENG and Y.Nakano,A precise classifier of the substances in urinary sediment images based on neural networks and fuzzy reasoning,IEEE2000
[49]L.A.Zadeh.A Fuzzy-Set-Theoretic Interpretation of Linguistic Hedges.Journal of Cybernetics 1972:2(2)
[50]何清,李洪兴.模糊聚类中的模糊等价矩阵[J].系统工程理论与实践,1999
[51]李相镐等.模糊聚类分析及其应用[M].贵阳:贵州科技出版社,1994
[52]Asultan K S,Selim S.A global algorithm for the fuzzy clustering problem[J].Pattern Recognition,1993(9)
[53]Fuguoyao.An Algorithm for Calculating the Global Optimal Fuzzy Equivalent Matrix of a Fuzzy Similarity Matrix[J].Fuzzy sets and system,1997(2)
[54]Bezdek J C.Clustering validity with fuzzy sets.Mathematical Biology,1907(I)
[55]Dave R N,Krishnapuram R.Robust clustering methods:A unified view.IEEE Fuzzy Systems,1997(2)
[56]何清,李洪兴.传递闭包与最优模糊等价矩阵[J].北京师范大学学报,1999(2)
[2]张勇,张强,虞烈.真彩色血细胞显微图像自动识别系统研究[J].西安交通大学学报,1999.33(2)
[3]刘圆圆.医学显微图像工作站[M].浙江:浙江大学通信与信息系统,2004
[4]吕小龙.尿沉渣有形成分自动分类识别算法研究[D].南京,东南大学生物医学工程系,2006.5
[5]李庆,韩红梅,耿鑫金.全自动尿沉渣检验误诊的综合因素[J].中国误诊学杂志,2002.10(2)
[6]王昌富,丛玉隆.尿液有形成分流式细胞分析技术及应用策略[J].国外医学临床生物化学与检验学分册,2003.24(2)
[7]陈红兵.尿液分析仪检测假阴性与假阳性原因探讨[J]浙江临床医学,2002.12(4)
[8]蒋红兵等.计算机显微图像尿沉渣分析仪的研究[J].医疗卫生装备,2001.4
[9]王选贺.基于尿沉渣显微图像的模式识别[D].长春,吉林大学,2007.4
[10]杜莉莉.基于分水岭算法的尿沉渣有形成分自动分类识别研究[D].长春,吉林大学,2007.4
[11]顾可梁.尿有形成分的识别与检查方法的选择[J].中华检验医学杂志,2005.28(6)
[12]Schanze T.Sinc Interpolation of discrete periodic signals.IEEE Transactions on Signal Processing,1999.47(11)
[13]马骏龙,丛玉隆,陈淑云等.全自动尿沉渣分析仪检测尿红细胞鉴别血尿来源的探讨[J].中华医学检验杂志,1997.20(5)
[14]顾可梁.尿沉渣镜检[J].上海医学检验杂志,2000.15(2)
[15]李思改,李卫,江志兰.尿液有形成分检查.尿化学分析法,尿人工显微镜检查UF-100尿沉渣全自动分析仪的比较[J].广东山市第二人民医院检验科,2001.12(2)
[16]尿显微镜:它的临床价值,澳洲悉尼皇室北海滨医院大学,内科、肾科,Sysmes Journal International Vol.6 No.1 1996
[17]谢华.基于多分类器融合的骨髓细胞识别技术研究[J].浙江大学硕士生毕业论文,2005
[18]蔡永军,刘伟玲,虞启琏.遗传神经网络在尿沉渣识别中的应用[J].医疗卫生装备,2004.1
[19]Beni C,Liu X M.A least biased fuzzy clustering method.IEEE Trans.PAMI1992,16(9)
[20]苗雪兰,刘瑞新,宋歌.计算机图形学理论及应用技术[M].北京:机械工业出版社,2007.2
[21]Zhang D S,Lu G J.Segmentation of moving objects in image sequences:Intelligence,2003.12(5)
[22]Kayargadde V,Martens J B.Estimation of edge parameters and image blur using polynomial transforms.CVGIP-GMIP,1996.56(6)
[23]Castleman K R.Digital Image Processing:Prentice-Hall,1996
[24]Cross G,Jain A.Markov random field texture models.IEEE Transactions on Pattern Analysis and Machine Intelligence,1983.5(1)
[25]Milan Sonka,Vaclav Hlavac,Roger Boyle.图像处理、分析与机器视觉[M].北京:人民邮电出版社,2003
[26]赵峰,赵荣椿.纹理分割及特征提取方法综述[J].中国体视学与图像分析,1998.3(4)
[27]钟珞等.模式识别[M].武汉:武汉大学出版社,2006
[28]基于神经网络的尿沉渣有形成分自动分类识别研究[D].杭州:浙江大学,2006.5
[29]章毓晋.图像工程(中)图像分析[M].北京:清华大学出版社,2005.10
[30]Zadeh.L A.Fuzzy sets[J].Information and Control,1965.8
[31]Zadeh.LA.Outline of a new approach to the analysis of complex systems and decision processes,IEEE Trans.Syst.,Man,Cybern,vol.SMC-3,1973
[32]刘普寅,吴孟达,模糊理论及其应用,长沙:国防科技大学出版社,2001.12
[33]何新贵,模糊知识处理的理论与技术,北京:国防工业出版社,1999.7
[34]L.A.Zadeh著,模糊集合-语言变量及模糊逻辑,北京:科学出版社,1982.5
[35]Ostu N.A,Threshold Selection Method from Gray-level Histogram,IEEE trans System Man Cybernetic,1978(8)
[36]夏良正.数字图象处理[M].南京:东南大学出版社,1999
[37]飞思科技产品研发中心,.Visual C++.NET编程指南[M].北京:电子工业出版社.2003,2-10
[38]K.R.Castleman著.朱志刚,石定机等译.数字图像处理.电子工业版,1998
[39]S.Arivazhagan,L.Ganesan.Texture segmentation using wavelet transform.Pattern recognition Letters,2003,24(23)
[40]Vincent Barra.Robust segmentation and analysis of DNA microarray spots using an adaptative split and merge algorithm.Computer Methods and Programs in Biomedicine,2006,81(2)
[41]边肇祺.张学工等,模式识别[M].北京:清华大学出版社.2000
[42]周长发.精通VisualC++图像编程[M].北京:电子工业出版社.2000
[43]Rafael C.Gonzalez Richard E.Woods.Digital Image Processing Second Edition[M].Beijing:Publishing House of Electronics Industry,2002.7
[44]W.Smith,The Scientist and Engineer's and Guide to Digital Signal Processing,California Technical Publishing,1997
[45]Haralick R M,Shapiro L G.Image segmentation techniques.Computer Vision,Graphics and Image Processing,1985
[46]Kittler J,Illingworth J,Foglein J.Threshold selection based on a simple image statistics.Computer Vision,Graphics and Image Processing,1985:30
[47]W.Ying,Y.Nakano,O.Nagata"A Method of Classifying Blood Cells in Urinary Sediments using Neural Network",Medical Imaging Technology,1998:2
[48]Ning Feng ZENG and Y.Nakano,A precise classifier of the substances in urinary sediment images based on neural networks and fuzzy reasoning,IEEE2000
[49]L.A.Zadeh.A Fuzzy-Set-Theoretic Interpretation of Linguistic Hedges.Journal of Cybernetics 1972:2(2)
[50]何清,李洪兴.模糊聚类中的模糊等价矩阵[J].系统工程理论与实践,1999
[51]李相镐等.模糊聚类分析及其应用[M].贵阳:贵州科技出版社,1994
[52]Asultan K S,Selim S.A global algorithm for the fuzzy clustering problem[J].Pattern Recognition,1993(9)
[53]Fuguoyao.An Algorithm for Calculating the Global Optimal Fuzzy Equivalent Matrix of a Fuzzy Similarity Matrix[J].Fuzzy sets and system,1997(2)
[54]Bezdek J C.Clustering validity with fuzzy sets.Mathematical Biology,1907(I)
[55]Dave R N,Krishnapuram R.Robust clustering methods:A unified view.IEEE Fuzzy Systems,1997(2)
[56]何清,李洪兴.传递闭包与最优模糊等价矩阵[J].北京师范大学学报,1999(2)
© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号 地址：北京市海淀区学院路29号 邮编：100083 电话：办公室：(+86 10)66554848；文献借阅、咨询服务、科技查新：66554700 |