Volume 8 Supplement 1
Automated classification of breast cancer morphology in histopathological images
© Ojansivu et al; licensee BioMed Central Ltd. 2013
Published: 30 September 2013
The morphology of a breast cancer tumour, as examined through an optical microscope, is currently assessed visually by the pathologist in parallel with making the cancer diagnosis. The grade of differentiation, which describes how closely the morphology of the tumour resembles the corresponding healthy tissue of an organ, is undisputedly related to the outcome of breast cancer . However, tumour grade is largely regarded as an unreliable prognostic factor due to its poor reproducibility . The visually determined morphology is afflicted with a poor inter- and intra observer agreement, which prevents grade from being fully utilized as an important outcome predictor. The same pathologist may assign different grade to the same tumour when assessment is repeated, and different pathologists disagree to a substantial level when assessing the same tumour .
Computational diagnostic tools for estimating the morphological properties of cancer tissue would enable objective and reproducible alternative for diagnosis. This could be achieved by fully utilizing the recent advances in digital microscopy and computer vision [4, 5]. Some attempts have already been made for automated grading of histopathological breast cancer images, but these studies have covered only limited amount of data or produce just a partial grading [6, 7]. We propose a texture based algorithm for automated classification of breast cancer morphology. The method uses the recently introduced LPQ  as well as LBP  descriptors and an SVM classifier. The LPQ and LBP descriptors each form a histogram representing the statistical texture properties and have been used earlier in many texture analysis applications which include surface inspection , tissue analysis , and face recognition , whereas SVM represents the state of the art among supervised learning based classification algorithms.
Material and methods
The images were transformed to gray scale and represented by LBP  and LPQ  texture descriptors. The classification of the images into the three classes was done using three one-versus-rest SVM classifiers with a radial basis function kernel (RBF) combined with chi-square distance metric. The final class was chosen by selecting the largest of the scores produced by the individual SVM classifiers. Given the training samples and their classes, an SVM classifier learns a model for the data which aims to separate the classes in space with a margin. In testing phase, the SVM classifier assigns new data samples into the classes based on the learned model. In our experiments, the data was split into two halves for training and testing of the SVM classifiers. We did additional experiments with only the extreme class 1 and 3 samples. In this case, we used the same descriptors and a binary SVM classifier with an RBF kernel.
Results and discussion
Histological grade of breast cancer is regarded as an important prognostic factor, but not included in staging guidelines due to the subjective nature of the assessment process. In the current study, we propose a computer vision method based on texture features and a classifier utilizing supervised machine learning to discriminate between cancer morphology as determined by a human observer. The results obtained show that automated grading is feasible and that discrimination between different levels of tubule formation can be performed with moderate to high accuracy. By combining LBP and LPQ features it is possible to improve the discrimination accuracy compared to using only LBP alone. While the extreme morphological structures according to tubule formation in the breast cancer tissue are discriminated with high accuracy, the recognition of the intermediate class should still be improved.
List of abbreviations used
Area under the ROC curve
Local binary pattern
Local phase quantization
Radial basis function
Receiver operating characteristic
Support vector machine
The study was kindly supported by the national Biomedinfra and Biocenter Finland projects.
- Elston CW, Ellis IO: Pathologic prognostic factors in breast cancer. I. The value of histological grades in breast cancer. Experience from a large study with long-term follow-up. Histopathology. 1991, 19: 403-410. 10.1111/j.1365-2559.1991.tb00229.x.View ArticlePubMedGoogle Scholar
- Singletary SE, Allred C, Ashley P, Bassett LW, Berry D, Bland KI, Borgen PI, Clark G, Edge SB, Hayes DF, Hughes LL, Hutter RVP, Morrow M, Page DL, Recht A, Theriault RL, Thor A, Weaver DL, Wieand HS, Greene FL: Revision of the American Joint Committee on Cancer staging system for breast cancer. J Clin Oncol. 2002, 20 (17): 3628-36. 10.1200/JCO.2002.02.026.View ArticlePubMedGoogle Scholar
- Boiesen P, Bendahl PO, Anagnostaki L, Domanski H, Holm E, Idvall I, Johansson S, Ljungberg O, Ringberg A, Ostberg G, Fernö M: Histologic grading in breast cancer: reproducibility between seven pathologic departments. Acta Oncol. 2000, South Sweden Breast Cancer Group, 39 (1): 41-5. 10.1080/028418600430950.
- Lundin M, Lundin J, Isola J: Virtual microscopy. J Clin Pathol. 2004, 57: 1250-1. 10.1136/jcp.2004.019919.PubMed CentralView ArticlePubMedGoogle Scholar
- Linder N, Konsti J, Turkki R, Rahtu E, Lundin M, Nordling S, Haglund C, Ahonen T, Pietikäinen M, Lundin J: Identification of tumor epithelium and stroma in tissue microarrays using texture analysis. Diagn Pathol. 2012, 7: 22-10.1186/1746-1596-7-22.PubMed CentralView ArticlePubMedGoogle Scholar
- Dalle JR, Leow WK, Racoceanu D, Tutac AE, Putti TC: Automatic breast cancer grading of histopathological images. Conf Proc IEEE Eng Med Biol Soc. 2008, 3052-5.Google Scholar
- Doyle S, Agner S, Madabhushi A, feldman M, Tomaszewski J: Automated grading of breast cancer histopathology using spectral clustering with textural and architectural image features. IEEE International Symposium on Biomedical Imaging: From Nano to Macro. 2008, 29: 496-499.Google Scholar
- Rahtu E, Heikkilä J, Ojansivu V, Ahonen T: Local Phase Quantization for Blur-Insensitive Image Analysis. Image and Vision Computing. 2012, doi: 10.1016/j.imavis.2012.04.001, Matlab code database [http://www.cse.oulu.fi/CMV/Downloads/LPQMatlab]Google Scholar
- Ojala T, Pietikäinen M, Mäenpää T: Multiresolution gray-scale and rotation invariant texture classification with local binary patterns. IEEE Transactions on Pattern Analysis and Machine Intelligence. 2002, 24 (7): 971-987. 10.1109/TPAMI.2002.1017623. Matlab code database [http://www.cse.oulu.fi/CMV/Downloads/LBPMatlab]View ArticleGoogle Scholar
- The FinProg breast cancer database. [http://www.finprog.org]
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.