A computer-aided diagnosis system of breast ductal lesion based on histopathological and imaging characteristics

doi:10.3969/j.issn.2095-4344.0328

Abstract

Abstract:

BACKGROUND: Histopathologic diagnosis of cells is an important method for the diagnosis of suspected breast cancer. If doctors need to use their own experience to judge each hematoxylin-eosin stained image with naked eyes, it is not only subjective, but also doctors’ fatigue and concentration affects the judgment results. At home and abroad, there is a lack of quantitative and objective computer-aided diagnosis system for breast cancer cells.
OBJECTIVE: To develop a computer-aided diagnosis system, based on the concept of multiple resolution, histopathological cytology and image structural characteristics, in order to identify the types of breast ductal hyperplasia.
METHODS: In order to introduce the concept of multi-resolution in the image of breast tube tissue of the patient’s thick needle puncture section, the cell image was first subjected to Sigmoid enhancement. Then the shape, size, arrangement and color uniformity of the cells were used as the discriminative features of the computer-aided diagnosis system, and the maximum expectation algorithm combined with the Lab color space were used for color separation of cell images, the cell nucleus of interest were preliminary segmented. The holes & cracks repair algorithm was used for the problem of incomplete nucleus division, and the watershed conversion was used for the problem of overlapping adjacent nuclear cells. Finally, the candidate cell nucleus was successfully isolated. The candidate nuclear feature acquisition was performed. Ellipse fitting, graph theory, and texture features were used to extract the required nuclear features, feature selection and linear discriminant analysis for feature acquisition and screening, and finally classified by the support vector machine. The device identifies the type of breast ductal cell hyperplasia to determine whether cancer activity exists.
RESULTS AND CONCLUSION: The experimental results of the interpretation of ductal hyperplasia confirmed that when using feature selection and linear discriminant analysis to obtain the 14-dimensional feature dimension and using support vector machine-radial basis function as a classifier, the accuracy of the diagnostic aid system designed in this paper was verified. With an accuracy of 88.4%, it can be used as an auxiliary system for the diagnosis of suspected breast cancer.

中国组织工程研究杂志出版内容重点：组织构建；骨细胞；软骨细胞；细胞培养；成纤维细胞；血管内皮细胞；骨质疏松；组织工程

Key words: Breast Neoplasms, Diagnosis, Computer-Assisted, Tissue Engineering

CLC Number:

R318

Tao Yong-peng, Liu Zhao-xia, Xu Cong. A computer-aided diagnosis system of breast ductal lesion based on histopathological and imaging characteristics[J]. Chinese Journal of Tissue Engineering Research, 2018, 22(24): 3893-3899.

Figures/Tables 3

2 乳腺导管增生的类型判别结果分析 Results 该系统的开发基于Windows 7-64bits操作系统，并采用Visual Studio 2008 Express C++，配合Open CV2.3与MATLab作为程序开发工具，得到乳腺导管细胞计算机辅助诊断系统。此文共选取了41张乳房导管细胞影像作为研究样本，其中普通导管细胞增生细胞影像23张，导管原位癌细胞影像18张。乳房导管细胞影像经患者同意后，取自瓦房店第三医院。前文中得到了细胞核的45个特征向量，但是如果45个特征向量全部使用，会导致影像特征维度较高，对计算能力和计算时间提出了较高的要求。因此，有必要对特征向量进行筛选。此文通过特征选择(Feature Selection)和线性判别分析进行特征向量的选择。最终，经特征选择筛选后特征向量为18维，经线性判别分析筛选后的特征向量为17维，而特征选择+线性判别分析筛选后的特征向量为14维。因此，此文采用4种特征维度结合支持向量机分类器对影像进行分类与比较，其中，4种特征维度包括：原始的全部45个特征维度、特征选择所筛选的18个特征维度、线性判别分析删选后的17个维度的特征，及特征选择配合线性判别分析筛选后的14个特征维度。对细胞核增生进行分类时，采用基于支持向量机分类模型，其主要参数设置kernel function核心为径向基函数核(RBF kernel)，迭代次数设置为25次，k-fold cross-validation设置为6(K=6)时得到最佳结果。表3为总共41张影像中，以不同判读方法分析分类结果。其中，病例影像1-23为普通导管细胞增生病例；24-41为导管原位癌病例。表中，打×表示以该方法判读时，会产生伪阳性的误判；打√表示以该方法判读时，会产生伪阴性的误判结果。比较判别结果可以发现，当采用的特征维度为45个时，错误的影像为14张；当采用的特征维度为特征选择选取的18个时，错误张数为12张；当采用特征维度为线性判别分析选取的17个时，其判读错误张数为9；"

References

[1] Mba HS, Ms SKG, Edwin Silverberg BS, et al. Survival experience in the breast cancer detection demonstration project. Ca A Cancer J Clin. 1987;37(5):258-290.

[2] 吕桂泉. 癌症不可怕: 30 年肿瘤诊治手记[M]. 杭州：浙江大学出版社, 2009.

[3] Poller DN, Barth A, Slamon DJ, et al. Prognostic classification of breast ductal carcinoma-in-situ. Lancet. 1995;345(8958): 1154-1157.

[4] 龚西騟. 有关乳腺导管上皮增生的几个问题[J]. 临床与实验病理学杂志, 1997,13(4):370-370.

[5] 辛智芳. 乳腺增生症的分类和诊治[J]. 中华乳腺病杂志(电子版), 2008, 2(6): 45-48.

[6] 许艳春,姚晓虹,祝明洁. 乳腺非典型性导管增生的研究进展[J]. 上海交通大学学报(医学版), 2013, 33(4):516-519.

[7] Dundar MM, Badve S, Bilgin G, et al. Computerized Classification of Intraductal Breast Lesions Using Histopathological Images. IEEE Trans Biomed Eng. 2011;58(7):1977-1984.

[8] Allison KH, Jensen KC. Intraductal Proliferations (DCIS, ADH, and UDH): A Comprehensive Guide to Core Needle Biopsies of the Breast. Springer International Publishing, 2016.

[9] Allred DC, Mohsin SK, Fuqua SA. Histological and biological evolution of human premalignant breast disease. Endocr Relat Canc. 2001;8(1):47-61.

[10] 杜辉,王树迎,徐淑敏. H.E.染色切片质量影响因素分析[J]. 中国兽医杂志,2003,39(12):50-51.

[11] Moon WK, Lo CM, Cho N, et al. Computer-aided diagnosis of breast masses using quantified BI-RADS findings. Comput Methods Programs Biomed. 2013;111(1):84-92.

[12] Basavanhally A, Ganesan S, Feldman M, et al. Multi-Field-of-View Framework for Distinguishing Tumor Grade in ER+ Breast Cancer From Entire Histopathology Slides. IEEE Trans Biomed Eng. 2013;60(8):2089-2099.

[13] 江贵平,秦文健,周寿军,等. 医学图像分割及其发展现状[J]. 计算机学报,2015,38(6):1222-1242.

[14] 缪仁拉,郑馨,江伟. 基于Lab 空间的白细胞图像偏色校正算法[J]. 电子世界,2017,39(15):153-153.

[15] 蔡隽,鲍旭东,吴磊,等. 基于活动轮廓模型的彩色白细胞图像自动分割方法研究[J]. 生物医学工程研究, 2005,24(4):218-222.

[16] 张艳玲,何鑫驰,李立. 基于最大类间方差与形态学的淋巴结图像分割[J]. 计算机科学,2013,40(8):296-299.

[17] 杨俊,郑曲波,吴桂良,等. 基于 Metropolis-SA 算法的脑部磁共振血管造影图像分割[J]. 生物医学工程与临床, 2013, 17(2): 113-118.

[18] 冯秀华,左玉霞,唐爱. 彩色多普勒检查乳腺疾病筛选早期乳癌的分析[J]. 医学影像学杂志,2003,13(6):399-401.

[19] 王蜀,李永宁,陈楷民,等. 基于数学形态学的医学图像分割[J]. 计算机应用,2005,25(10):2381-2382.

[20] 赵于前. 基于数学形态学的医学图像处理理论与方法研究[M]. 中南大学, 2006.

[21] 夏剑峰. 基于数学形态学的癌细胞的分割与识别[J].电子科技, 2016, 29(10): 11.

[22] 马东, 曹培杰. 分割重叠细胞核的方法及比较研究[J]. 北京生物医学工程, 1999, 18(3): 142-147.

[23] Mouelhi A, Sayadi M, Fnaiech F, et al. Automatic image segmentation of nuclear stained breast tissue sections using color active contour model and an improved watershed method. Biomed Sign Proc Contr. 2013;8(5):421-436.

[24] 胡彩亮. 基于指纹图的白细胞图像分割算法研究[D]. 燕山大学, 2011.

[25] Vincent L, Soille P. Watersheds in Digital Spaces: An Efficient Algorithm Based on Immersion Simulations. IEEE Trans Patt Anal Mach Intell. 1991;13(6):583-598.

[26] 廖苗,赵于前,曾业战,等. 基于支持向量机和椭圆拟合的细胞图像自动分割[J]. 浙江大学学报(工学版), 2017, 51(4):722-728.

[27] 王康,李晓春. 基于多空间图像融合的白细胞自动分割[J]. 计算机仿真, 2015, 32(3): 258-262.

[28] Caselles V, Kimmel R, Sapiro G. Geodesic active contours. Int J Comp Vis. 1997;22(1): 61-79.

[29] Pham VT, Tran TT, Chiu YJ, et al. Region-aided Geodesic Active Contour model for image segmentation. IEEE International Conference on Computer Science and Information Technology. IEEE, 2010:318-321.

[30] Aubert G, Barlaud M, Faugeras O, et al. Image segmentation using active contours: Calculus of variations or shape gradients? SIAM J Appl Math. 2003;63(6): 2128-2154.

[31] Fatakdawala H, Xu J, Basavanhally A, et al. Expectation-maximization-driven geodesic active contour with overlap resolution (emagacor): Application to lymphocyte segmentation on breast cancer histopathology. IEEE Trans Biomed Eng. 2010;57(7):1676-1689.

[32] 龚平,郭华雄,王文清,等. 基于人工神经网络的肺癌细胞形态学诊断模型的建立及应用[J]. 中华生物医学工程杂志, 2012,18(1):73-76.

[33] 董敏,马义德. 基于乳腺肿瘤细胞形态特征参数的乳腺癌诊断发展研究[J]. 中华临床医师杂志:电子版, 2013, 7(11):5023-5026.

[34] 龚西騟. 乳腺活检中热点问题的新认识(一)[J]. 临床与实验病理学杂志, 2005, 21(1):1-5.

[35] 姜英,陈杰. 乳腺导管上皮内瘤变(DIN)[J]. 诊断病理学杂志, 2005, 12(6):463-466.

[36] 马闪闪,董明利,张帆,等. 基于核主成分分析的流式细胞数据分群方法研究[J]. 生物医学工程学杂志, 2017,34(1):115-122.

[37] 赵明珠. 细胞病理图像的特征分析与分类识别[D]. 浙江工业大学, 2012.

[38] Soh LK, Tsatsoulis C. Texture analysis of SAR sea ice imagery using gray level co-occurrence matrices. IEEE Trans Geosc Rem Sens. 1999;37(2): 780-795.

[39] Yu J. Texture Image Segmentation Based on Gaussian Mixture Models and Gray Level Co-occurrence Matrix[C]. International Symposium on Information Science and Engineering. IEEE, 2010:149-152.

[40] 白雪冰,王克奇,王辉. 基于灰度共生矩阵的木材纹理分类方法的研究[J]. 哈尔滨工业大学学报, 2005, 37(12):1667-1670.