Novel Texture Feature Descriptors Based on Multi-Fractal Analysis and LBP for Classifying Breast Density in Mammograms

Abstract

This paper investigates the usefulness of multi-fractal analysis and local binary patterns (LBP) as texture descriptors for classifying mammogram images into different breast density categories. Multi-fractal analysis is also used in the pre-processing step to segment the region of interest (ROI). We use four multi-fractal measures and the LBP method to extract texture features, and to compare their classification performance in experiments. In addition, a feature descriptor combining multi-fractal features and multi-resolution LBP (MLBP) features is proposed and evaluated in this study to improve classification accuracy. An autoencoder network and principal component analysis (PCA) are used for reducing feature redundancy in the classification model. A full field digital mammogram (FFDM) dataset, INBreast, which contains 409 mammogram images, is used in our experiment. BI-RADS density labels given by radiologists are used as the ground truth to evaluate the classification results using the proposed methods. Experimental results show that the proposed feature descriptor based on multi-fractal features and LBP result in higher classification accuracy than using individual texture feature sets.

Elimination of Defects in Mammograms Caused by a Malfunction of the Device Matrix

“…Binarization is also actively used in mammography to eliminate image noise [ 32 , 33 ]. Other approaches are also used to highlight an image against the background; for example, the thresholding method and morphological operations were effectively applied in [ 17 ].…”

“…It was proposed the use of the following procedures to improve the quality: low-pass filter, Gaussian smoothing, subsampling operations, and morphological operations. To eliminate noise, median filtering was also used with a 3 by 3 window [ 16 , 17 ]. In [ 18 ], various image quality improvement algorithms were presented, such as the synthetic minority over-sampling technique to improve the training set; a Gaussian pyramid for image scaling with minimal loss; histogram equalization, an adaptive mean, median filters, log transforms, and a Wiener filter to increase contrast; and thresholding of the pixel intensity to eliminate artifacts.…”

A Genetic Programming Approach to Automatically Construct Informative Attributes for Mammographic Density Classification

“…Though the author performed thorough investigation to increase the classification performance, the method is not interpretable and cannot identify important attributes necessary to discriminate between breast density classes. Li et al [2] proposed a multi-step process including breast region segmentation, feature extraction using local Binary patterns (LBP) and multi-texture fractal features, feature selection using principal component analysis and an autoencoder, and MBD classification using SVM. Their method is tested on only one dataset and all images are resized to a single resolution which distorts the aspect-ratio resulting in loss of texture information.…”

“…1 shows the incidence and mortality rates of the top 10 most common cancers where breast cancer remains prominent. Breast density is an important indicator for developing breast cancer [2]. In medical studies, a woman gets 5-fold increased risk of developing breast cancer if mammographic breast density (MBD) exceeds 75% [3].…”

“…The standard BI-RADS criterion has four categories; BI-RADS I: fatty (0-25%), BI-RADSII: scattered density (26-50%), BI-RADSIII: heterogeneously dense (51-75%), and BI-RADSIV: extremely dense (76-100%) [5]. When radiologists observe mammograms with the naked eye to classify them into BI-RADS categories, there is an obvious inter-and intra-reader variability [2]. In addition, categorizing a large number of mammograms is a tiresome process for the radiologists.…”