Robust Segmentation, Shape Fitting and Morphology Computation of High-Throughput Cell Nuclei

doi:10.1007/s12204-017-1819-3

Abstract

Abstract: Accurate nuclear classification (e.g., grading of renal cell carcinoma (RCC) biopsy images) is important to better understand fundamental phenomena such as tumor growth. In this paper, an automated pipeline is proposed to quantitatively analyze RCC data. A novel segmentation methodology is firstly used to delineate cell nuclei based on minimum description length (MDL) constrained B-spline curve fitting. From the obtained segmentations, thirteen features are then extracted based on five types of characteristics. These features are used to classify cell nuclei in biopsy images. Associations among nuclei are computed and represented by graphical networks to enable further analysis. Finally, a support vector machine (SVM) based decision-graph classifier is introduced to classify the biopsy images with the purpose of grading. Experimental results on real RCC data show that our SVM-based decision-graph classifier achieves 95.20% of classification accuracy while the SVM classifiers achieve 93.33% of classification accuracy.

摘要： Accurate nuclear classification (e.g., grading of renal cell carcinoma (RCC) biopsy images) is important to better understand fundamental phenomena such as tumor growth. In this paper, an automated pipeline is proposed to quantitatively analyze RCC data. A novel segmentation methodology is firstly used to delineate cell nuclei based on minimum description length (MDL) constrained B-spline curve fitting. From the obtained segmentations, thirteen features are then extracted based on five types of characteristics. These features are used to classify cell nuclei in biopsy images. Associations among nuclei are computed and represented by graphical networks to enable further analysis. Finally, a support vector machine (SVM) based decision-graph classifier is introduced to classify the biopsy images with the purpose of grading. Experimental results on real RCC data show that our SVM-based decision-graph classifier achieves 95.20% of classification accuracy while the SVM classifiers achieve 93.33% of classification accuracy.

CLC Number:

TP 391.4

SONG Jie (宋杰), XIAO Liang* (肖亮), LIAN Zhichao (练智超). Robust Segmentation, Shape Fitting and Morphology Computation of High-Throughput Cell Nuclei[J]. Journal of shanghai Jiaotong University (Science), 2017, 22(2): 180-187.

References 17

[1]	COSATTO E, MILLER M, GRAF H P, et al. Gradingnuclear pleomorphism on histological micrographs[C]//Proceedings of International Conference on PatternRecognition. New York, USA: IEEE, 2008: 1-4.
[2]	CHEKKOURY A, KHURD P, NI J, et al. Automatedmalignancy detection in breast histopathological images[C]//Proceedings of SPIE Medical Imaging. SanDiego, CA, USA: SPIE, 2012: 831515-1-831515-13.
[3]	DOYLE S, AGNER S, MADABHUSHI A, et al. Automatedgrading of breast cancer histopathology usingspectral clustering with textural and architecturalimage features [C]//Proceedings of 5th IEEE InternationalSymposium on Biomedical Imaging. Paris,France: IEEE, 2008: 496-499.
[4]	BASAVANHALLY A, GANESAN S, FELDMAN M,et al. Multi-field-of-viewframework for distinguishingtumor grade in ER+ breast cancer from entirehistopathology slides [J]. IEEE Transactions onBiomedical Engineering, 2013, 60(8): 3052-3055.
[5]	LOHSE C M, BLUTE M L, ZINCKE H, et al. Comparisonof standardized and non-standardized nucleargrade of renal cell carcinoma to predict outcome among2042 patients [J]. American Journal of Clinical Pathology,2002, 118(6): 877-886.
[6]	KIM T Y, CHOI H J, CHA S J, et al. Study on textureanalysis of renal cell carcinoma nuclei based onthe Fuhrman grading System [C]//Proceedings of SeventhInternational Workshop on Enterprise Networkingand Computing in Healthcare Industry. Busan, Korea:IEEE, 2005: 384-387.
[7]	NOVARA G, MARTIGNONI G, ARTIBANI W, et al.Grading systems in renal cell carcinoma [J]. Journal ofUrology, 2007, 177(2): 430-436.
[8]	BJORNSSON C S, LIN G, AL-KOFAHI Y, et al. Associativeimage analysis: A method for automated quantificationof 3D multi-parameter images of brain tissue[J]. Journal of Neuroscience Methods, 2008, 170(1):165-178.
[9]	FIGUEIREDO M, LEITAO J, JAIN A. Unsupervisedcontour representation and estimation using B-splinesand a minimum description length criterion [J]. IEEETransactions on Image Processing, 2000, 9(6): 1075-1086.
[10]	AL-KOFAHI Y, LASSOUED W, LEE W, et al. Improvedautomatic detection and segmentation of cellnuclei in histopathology images [J]. IEEE Transactionson Biomedical Engineering, 2010, 57(4): 841-852.
[11]	HARALICK R M, SHANMUGAM K, DINSTEIN I.Texture features for image classification [J]. IEEETransactions on Systems, Man, and Cybernetics, 1973,SMC-3(6): 610-621.
[12]	SONG J, XIAO L, LIAN Z C. Boundary-to-marker evidencecontrolled segmentation and MDL-based contourinference for overlapping nuclei [J]. IEEE Journalof Biomedical and Health Informatics, 2015. DOI:10.1109/JBHI.2015. 2504422 (published online).
[13]	FLICKNER M, HAFNER J, RODRIGUEZ E, et al.Periodic quasi-orthogonal spline basis and applicationsto least squares over curve fitting of digital images [J].IEEE Transactions on Image Processing, 1996, 5(1):71-88.
[14]	GRUNWALD P, MYUNG J, PITT M. Advances inminimum description length: Theory and applications[M]. Cambridge: MIT Press, 2004.
[15]	LOLIVE D, BARBOT N, BOEFFARD O. Melodiccontour estimation with B-spline models using a MDLcriterion [C]//Proceedings of International Conferenceon Speech and Computer. St. Petersburg, Russia: AnatolyaPublishers, 2006: 333-338.
[16]	CHAM T J, CIPOLLA R. Automated B-splinecurve representation incorporating MDL and errorminimizingcontrol point insertion strategies [J]. IEEETransactions on Pattern Analysis & Machine Intelligence,1999, 21(1): 49-53.
[17]	FUHRMAN S A, LASKY L C, LIMAS C. Prognosticsignificance of morphologic parameters in renal cellcarcinoma [J]. American Journal of Surgical Pathology,1982, 6(7): 655-663.