Texture analysis of masses malignant in mammograms images using a combined approach of diversity index and local binary patterns distribution
Introduction
The World Health Organization (WHO) estimates that in 2015, at least 561 thousand women will die of breast cancer throughout the world. Despite being considered a disease of the developed world, nearly 50% of the cases and 58% of the deaths by this illness occur in less developed countries, mainly due to late diagnoses (WHO, 2013).
Cancer control is closely related to its precocious diagnosis. Early detection widens the range of therapeutic options, improving the patient’s life quality and raising the chances of a cure. It is a consensus that the mammogram is the most important exam to detect breast cancer in women after the age of 40. This exam allows the detection of cancer even at an initial stage, when the tumor is not yet palpable. Mammograms detect about 80% to 90% of cases of breast cancer without symptoms (ACS, 2011). However, some factors, such as the patient’s age, physician’s experience and quality of the images, may cause great variations in the sensitivity of this exam.
The use of image processing and machine learning techniques has contributed to the task of cancer detection, allowing a more precise diagnosis. Therefore, there has been an increased interest over the last decade in the development of a computer-aided detection and diagnosis system (CAD and CADx, respectively) that can help radiologists with the interpretation of mammographic exams (Kinoshita, Pereira, Honda, Rodrigue, & Marques, 2004).
By observing mammogram images, it is possible to detect a mass, which is one of the abnormalities visible in this type of exam. These masses are agglomerate cells that unite more densely than the tissues that surround them, and can be the result of both malignant and benign conditions. For this reason, information such as size, shape and location of the margins of a mass are very important to determine its probability of being malignant (Kopans, 2006).
Texture is a feature that is difficult for humans to analyze. They normally use the contour of the masses to diagnose these regions. However, such features are not always easily distinguishable in these exams. There are some lesions without a well-defined contour that hinder a correct visualization. This difficulty ends up increasing the number of biopsies with negative results. The development of techniques for extraction of features may help experts in providing a more precise diagnosis by supplying a second opinion.
These difficulties justify the need for the development of techniques that solely use texture analysis to extract features, so that mammogram images that do not present well-defined contours can be used efficiently to determine the probability of malignancy or benignity, thereby providing the expert with greater support in the diagnosis of breast cancer.
The purpose of this article is to propose a methodology solely based on texture analysis to describe the features of masses in mammograms. To this end, the methodology proposes the concept of patterns of diversity by using the co-occurrence of species, which is determined by the analysis of the Local Binary Patterns (LBPs) (Ojala, Pietikainen, & Harwood, 1996) that exist in these masses and applying diversity indexes to evaluate them. We use the Shannon, McIntosh, Simpson, Gleason and Menhinick indexes. Finally, the extracted feature is classified using a Support Vector Machine (SVM) in order to differentiate between malignant and benign masses.
Different shape and texture descriptors are proposed to classify masses detected within mammographic images as malignant or benign. A good classification is directly related to descriptor characteristics and the ability to discriminate the images in these two classes. Thus, researchers are always looking for descriptors that can be easily computed and that have the power to discriminate the classes, contributing considerably to the classification performance. Based on this, we believe that the work presented in the paper contributes to the expert system area in the following ways: it uses a cascade approach for structural and statistical representations aiming to obtain the texture pattern description; it adapts ecology indexes in a way that is able to describe texture features; and it is simple with a high discriminatory power and can easily be extended to other image applications.
The remainder of this article is organized in five sections. In Section 2, we describe some related works. Then, in Section 3, we give details of the proposed methodology. In Section 4, we present and discuss the results of the experiments carried out with the application of the proposed method. Finally, in Section 5, we establish some conclusions, as well as suggestions for future works.
Section snippets
Related works
Many works have been undertaken with the goal of analyzing texture in order to differentiate suspect regions in mammograms and suggest their malignant or benign behavior. A common strategy to extract texture features is the use of gray-level co-occurrence matrices (GLCMs). Several works adopted this technique (Naveed, Jaffar, T., 2011, Panda, Baig, Panigrahi, Patro, 2015, Pereira, Marques, Honda, Kinoshita, Engelmann, Muramatsu, Doi, 2007, Rangayyan, Nguyen, Ayres, Nandi, 2010, Vasantha,
Proposed methodology
In this section, we describe the procedures adopted in the proposed methodology for the differentiation of malignant and benign patterns in masses based on mammographic images. Fig. 1 presents the stages of the methodology, namely: image acquisition, preprocessing, ROI representation, feature extraction and pattern recognition.
Results and discussion
As stated in the description of the methodology, for each training/test proportion, we executed five repetitions. The parameters used for each test were estimated by cross validation. The parameters are unique for each test and represent the function that maps the feature vectors onto the support vectors. Therefore, owing to the random selection of the training and test bases, they cannot be reused.
We present the mean values and standard deviations for accuracy, sensitivity, specificity,
Conclusion
Shape is an important characteristic used by specialists to diagnose the malignancy of suspect regions on mammographic images. In general, shapes are classified using the Breast Imaging Reporting and Data System (BI-RADS) classifications, but due to breast tissue compression during the mammography image acquisition, the precise detection of mass shape in the image is a difficult task for image processing techniques.
In this paper, we present a methodology for characteristic extraction of mass
Acknowledgment
The authors acknowledge CAPES, CNPq and FAPEMA for their financial support.
References (54)
- et al.
Textural features for image classification
IEEE Transactions on Systems, Man and Cybernetics
(1973) An index of diversity and the relation of certain concepts to diversity
Ecological Society of America
(1967)A comparison of some species-individuals diversity indices applied to samples of field insects
Ecology
(1964)- et al.
MRT letter: segmentation and texture-based classification of breast mammogram images
Microscopy Research and Technique
(2011) - et al.
A comparative study of texture measures with classification based on feature distribuition
Pattern Recognition
(1996) - et al.
Usefulness of texture analysis for computerized classification of breast lesions on mammograms
Journal of Digital Imaging
(2007) - et al.
Improvement of mammographic mass characterization using spiculation measures and morphological features
Medical Physics
(2007) - et al.
An automatic and simple breast tumor classification using area matching
Image information processing (ICIIP), 2011 international conference on
(2011) Learn about breast cancer
(2011)- et al.
Classification of breast masses in digital mammograms using support vector machines
International Journal of Advanced Research in Computer Science and Software Engineering
(2013)
Texture analysis for mammographic breast images using novel statistical and transform techniques
(IJARTET) International Journal of Advanced Research Trends in Engineering and Technology
An introduction to medical statistics
Classification of breast tissues using moran’s index and geary’s coefficient as texture signatures and svm
Computers in Biology and Medicine
Field and laboratory methods for general ecology
Lbp operators on curvelet coefficients as an algorithm to describe texture in breast cancer tissues
Expert Systems with Applications
Diagnosis of breast masses from dynamic contrast-enhanced and diffusion-weighted MR: a machine learning approach
PLoS ONE
LIBSVM: a library for support vector machines
ACM Transactions on Intelligent Systems and Technology
Mammographic masses classification: novel and simple signal analysis method
Electronics Letters
Texture analysis using gray level run lengths
Computer Graphics and Image Processing
Digital image processing
Texture unit, texture spectrum, and texture analysis
IEEE Transactions on Geoscience and Remote Sensing
DDSM: the digital database for screening mammography
Proceedings of the 5th international workshop on digital mammography
An efficient automatic mass classification method in digitized mammograms using artificial neural network
International Journal of Artificial Intelligence and Applications
An automatic method for detection of the nipple and pectoral muscle in digitized mammograms
Congresso Latino-Americano de engenharia biomédica (CLAEB 2004)
Comparison of feature extraction algorithms for mammography images
Image and signal processing (CISP), 2011 4th international congress on
Breast imaging
Classification of mammographic masses using generalized dynamic fuzzy neural networks
Medical Physics
Cited by (21)
Automated breast cancer detection in mammography using ensemble classifier and feature weighting algorithms
2023, Expert Systems with ApplicationsA DenseNet CNN-based liver lesion prediction and classification for future medical diagnosis
2023, Scientific AfricanComputer-aided breast cancer detection and classification in mammography: A comprehensive review
2023, Computers in Biology and MedicineAnchor-free YOLOv3 for mass detection in mammogram
2022, Expert Systems with ApplicationsCitation Excerpt :According to statistics, breast cancer accounts for 22% of all newly diagnosed cancers, causing 13.7% cancer related deaths all over the world (Freddie et al., 2018). Fortunately, when detected early, breast cancer is one of the most treatable malignancies (Da Rocha et al., 2016; Swiderski et al., 2017). Among many imaging techniques, mammography is a well-known method for breast cancer detection (Baâzaoui et al., 2018; Pedro et al., 2019).
A review on image-based approaches for breast cancer detection, segmentation, and classification
2021, Expert Systems with ApplicationsCitation Excerpt :The tumor composition is classified using the Breast Imaging Reporting and Data System (BI-RADS) into category I (25%), category II (50% of), category III (50–75%), and category IV(75%) of glandular density (de Oliveira Silva, Barros, & Lopes, 2017). However due to breast tissue compression during the mammography image acquisition, the precise detection of mass shape in the image is a difficult task for image processing techniques (da Rocha et al., 2016). A free open source tool, namely CellProfiler (Carpenter et al., 2006) is applicable to pathological images and is developed to help the researchers in order to characterize the features in images.
Fuzzy C-means and region growing based classification of tumor from mammograms using hybrid texture feature
2018, Journal of Computational ScienceCitation Excerpt :LBP is another well-known texture-based description mechanism employed by several researchers. Bruno et al. [23], de Rocha et al. [24], Liu et al. [25] applied LBP to identify benign masses with respect to their malignancy. Singh et al. used center symmetric-LBP [26] and Nanni et al. [27] employed LTP (Local Ternary Pattern) and LPQ (Local Phase Quantization) as texture-based features.