Wenda He

A Review on Automatic Mammographic Density and Parenchymal Segmentation

Breast cancer is the most frequently diagnosed cancer in women. However, the exact cause(s) of breast cancer still remains unknown. Early detection, precise identification of women at risk, and application of appropriate disease prevention measures are by far the most effective way to tackle breast cancer. There are more than 70 common genetic susceptibility factors included in the current non-image-based risk prediction models (e.g., the Gail and the Tyrer-Cuzick models). Image-based risk factors, such as mammographic densities and parenchymal patterns, have been established as biomarkers but have not been fully incorporated in the risk prediction models used for risk stratification in screening and/or measuring responsiveness to preventive approaches. Within computer aided mammography, automatic mammographic tissue segmentation methods have been developed for estimation of breast tissue composition to facilitate mammographic risk assessment. This paper presents a comprehensive review of automatic mammographic tissue segmentation methodologies developed over the past two decades and the evidence for risk assessment/density classification using segmentation. The aim of this review is to analyse how engineering advances have progressed and the impact automatic mammographic tissue segmentation has in a clinical environment, as well as to understand the current research gaps with respect to the incorporation of image-based risk factors in non-image-based risk prediction models.

Mammographic segmentation and risk classification using a novel binary model based bayes classifier

Clinical research has shown that the sensitivity of mammography is significantly reduced by increased breast density, which can mask some tumours due to dense fibroglandular tissue. In addition, there is a clear correlation between the overall breast density and mammographic risk. We present an automatic mammographic density segmentation approach using a novel binary model based Bayes classifier. The Mammographic Image Analysis Society (MIAS) database was used in a quantitative and qualitative evaluation. Visual assessment on the segmentation results indicated a good and consistent extraction of mammographic density. With respect to mammographic risk classification, substantial agreements were found between the classification results and ground truth provided by expert screening radiologists. Classification accuracies were 85% and 78% in Tabar and Breast Imaging Reporting and Data System (Birads) categories, respectively; whilst in the corresponding low and high categories, the classification accuracies were 93% and 88% for Tabar and Birads, respectively.

Mammographic image segmentation and risk classification based on mammographic parenchymal patterns and geometric moments

Abstract Mammographic risk assessment is becoming increasingly important in decision making in screening mammography and computer aided diagnosis systems. Strong evidence shows that characteristic patterns of breast tissue as seen in mammography, referred to as mammographic parenchymal patterns, provide crucial information about breast cancer risk. Quantitative evaluation of the characteristic mixture of breast tissue can be used both for the estimation of mammographic risk assessment, as well as for quantifying the change of the relative proportion of different breast tissue patterns. This paper investigates mammographic image segmentation based on geometric moments, and prior information of the mammographic building blocks as described by Tabar tissue modelling. The segmentation methodology presented here consists of five distinct steps: (1) feature extraction using mammographic patches, (2) deriving local image properties, (3) feature transformation, (4) mammographic building block based model generation by clustering, and (5) model driven segmentation. The Mammographic Image Analysis Society database was used to facilitate the quantitative and qualitative evaluation, with respect to mammographic risk assessment, based on both Tabar and Breast Imaging Reporting And Data System schemes. Classification accuracies of 71% and 79% were achieved in the corresponding low and high risk categories for Tabar and Breast Imaging Reporting And Data System schemes, respectively. Visual assessment indicates that the proposed segmentation approach can produce consistent and realistic segmentation results, with respect to breast anatomy and Tabar tissue modelling. For screening mammography and computer aided diagnosis, the proposed mammographic segmentation approach is useful in aiding radiologists’ estimation of breast cancer risk and treatment planning prior to biopsies.

Mammographic Segmentation Based on Texture Modelling of Tabár Mammographic Building Blocks

We present an approach to automate texton selection to achieve optimized mammogram segmentation results with respect to mammographic building blocks (i.e. nodular, linear, homogeneous, and radiolucent) as described by Tabars tissue model. Such segmentation results are expected to lead to improvements in automatic mammographic risk assessment modelling. The texton selection process has three distinct components, covering a) texton ranking, b) outlier detection, and c) visual assessment. The initial results, on tissue specific regions and full mammographic images are promising, but at the same time indicate shortcomings, which are discussed.

Mammographic segmentation based on mammographic parenchymal patterns and spatial moments

Strong evidence shows that characteristic patterns of breast tissues as seen on mammography, referred to as mammographic parenchymal patterns, provide crucial information about breast cancer risk. Quantitative evaluation of the characteristic mixture of breast tissues can be used as for mammographic risk assessment as well as for quantification of change of the relative proportion of different breast tissue patterns. This paper investigates mammographic segmentation based on spatial moments and prior information of mammographic building blocks (i.e. nodular, linear, homogenous, and radiolucent) as described by Tabárs tissue models to describe parenchymal patterns. The algorithm extracted texture features from a set of sub-sampled mammographic patches. Tabárs mammographic building blocks were modelled as statistical distribution of clustered filter responses based on spatial moments. Evaluation was based on the Mammographic Image Analysis Society (MIAS) database. The experimental results indicated that the developed methodology is capable of modelling complex mammographic images and can deal with intraclass variation and noise aspects. The results show realistic segmentation on tissue specific regions with respect to breast anatomy and Tabárs tissue models. In addition, the segmentation results were used for mammographic risk based classification of the entire MIAS database resulting in ∼70% correct low/high risk classification.

Breast image pre-processing for mammographic tissue segmentation

During mammographic image acquisition, a compression paddle is used to even the breast thickness in order to obtain optimal image quality. Clinical observation has indicated that some mammograms may exhibit abrupt intensity change and low visibility of tissue structures in the breast peripheral areas. Such appearance discrepancies can affect image interpretation and may not be desirable for computer aided mammography, leading to incorrect diagnosis and/or detection which can have a negative impact on sensitivity and specificity of screening mammography. This paper describes a novel mammographic image pre-processing method to improve image quality for analysis. An image selection process is incorporated to better target problematic images. The processed images show improved mammographic appearances not only in the breast periphery but also across the mammograms. Mammographic segmentation and risk/density classification were performed to facilitate a quantitative and qualitative evaluation. When using the processed images, the results indicated more anatomically correct segmentation in tissue specific areas, and subsequently better classification accuracies were achieved. Visual assessments were conducted in a clinical environment to determine the quality of the processed images and the resultant segmentation. The developed method has shown promising results. It is expected to be useful in early breast cancer detection, risk-stratified screening, and aiding radiologists in the process of decision making prior to surgery and/or treatment.

A Study on Mammographic Image Modelling and Classification Using Multiple Databases

Within computer aided mammography, there are many image analysis methods have been developed for mammographic image classification. Some of these were developed and validated using well known publicly available databases, and others may have chosen to use independent/private databases for their investigations. Often, despite the promising results described in the literature, it is not unusual to see when adapting an established method with the recommended configurations for a different database, the obtained results are not in line with expectation. This paper presents results of a study with respect to the implications of mammographic image classification using different classifiers trained with variations, such as differences in parameter settings, classifiers, using single databases, combined and across databases. The results indicated that it is unlikely to have an universal parameter settings and classifiers, which can be used to achieve the best classification without tuning. Additional databases used at the training stages do not necessarily lead to more accurate density classifications; whilst classifiers trained with images obtained using one type of image acquisition are not ideal for classifying images obtained using different image acquisition. The related issues of optimal parameter configuration, classifier selection, and utilising single or multiple databases at the training stage are discussed.

Combining Texture and Hyperspectral Information for the Classification of Tree Species in Australian Savanna Woodlands

This paper outlines research undertaken to assess the ability of textural information, from image filters, to be used alongside hyperspectral data for the classification of broad forest types. The study made use of 2.6 m hyperspectral HyMap data acquired over the Injune study area, Queensland, Australia, in September 2000. The HyMap data provided spectral data from the blue to shortwave infrared in 126 wavelengths, all of which were used for classification. A measure of texture was achieved using a set of 48 image filters including Laplacian of Guassian and Gaussian smoothing, first and second order derivatives at different scale and where appropriate different rotations. Analysis took place using an air photo interpretation to provide regions of interest for areas dominated by Angophora, Callitris, and Eucalyptus, additionally areas of non-forest were also included. Classification of the resulting dataset was performed using Multiple Stepwise Discriminant Analysis where an accuracy of 60% was achieved using the combined reflectance and texture data compared to accuracies of 55 and 43% using only the reflectance and textural datasets, respectively.

Mammographic Segmentation and Density Classification: A Fractal Inspired Approach

Breast cancer is the most frequently diagnosed cancer in women. To date, the exact causes of breast cancer still remains unknown. The most effective way to tackle the disease is early detection through breast screening programmes. Breast density is a well established image based risk factor. An accurate dense breast tissue segmentation can play a vital role in precise identification of women at risk, and determining appropriate measures for disease prevention. Fractal techniques have been used in many biomedical image processing applications with varying degrees of success. This paper describes a fractal inspired approach to mammographic tissue segmentation. A multiresolution stack representation and 3D histogram features extended from 2D are proposed. Quantitative and qualitative evaluation was performed including mammographic tissue segmentation and density classification. Results showed that the developed methodology was able to differentiate between breast tissue variations. The achieved density classification accuracy for 360 digital mammograms is 78i¾?% based on the BI-RADS scheme. The developed fractal inspired approach in conjunction with the stack representation and 3D histogram features has demonstrated an ability to produce quality mammographic tissue segmentation. This in turn can be found useful in early breast cancer detection, risk-stratified screening, and aiding radiologists in the process of decision making prior to surgery and/or treatment.

A Revisit on Correlation between Tabár and Birads Based Risk Assessment Schemes with Full Field Digital Mammography

Mammographic risk assessment is used to determine the probability of a woman developing breast cancer and it plays an important role in the early detection and disease prevention within screening mammography. Tabar and Birads are two fundamentally different risk schemes, one is assessed based on mixtures of breast parenchyma and the other one is assessed based on the percentage of dense breast tissue. This paper presents findings on the correlation between these two mammographic risk assessment schemes; aspects with respect to reader experience and related inter reader variability were also investigated. As a follow up (revisit) investigation to a previously published paper, the new results have shown a strong correlation between Tabar and Birads with the highest Spearman’s correlation coefficient > 0.92 and κ = 0.86% (almost perfect agreement). The statistical results vary with readers’ mammographic reading experience, which also indicated subtle information such as that some mixture of breast parenchma (Tabar specific mammographic building blocks) may be more likely to cause inter reader variability.