Elsevier

Applied Soft Computing

Volume 35, October 2015, Pages 175-185
Applied Soft Computing

Mammogram image visual enhancement, mass segmentation and classification

https://doi.org/10.1016/j.asoc.2015.06.029Get rights and content

Highlights

  • Reveal the optimal combination of various enhancement methods.

  • Segment breast region in order to obtain better visual interpretation.

  • To assist radiologists in making accurate decisions, analysis and classifications.

  • Tumor classification accuracy and sensitivity values of 81.1% and 86%, respectively.

  • Participated radiologists are pleased with the results and acknowledged the work.

Abstract

Mammography is the most effective technique for breast cancer screening and detection of abnormalities. However, early detection of breast cancer is dependent on both the radiologist's ability to read mammograms and the quality of mammogram images. In this paper, the researchers have investigated combining several image enhancement algorithms to enhance the performance of breast-region segmentation. The masses that appear in mammogram images are further analyzed and classified into four categories that include: benign, probable benign and possible malignant, probable malignant and possible benign, and malignant. The main contribution of this work is to reveal the optimal combination of various enhancement methods and to segment breast region in order to obtain better visual interpretation, analysis, and classification of mammogram masses to assist radiologists in making more accurate decisions. The experimental dataset consists of a total of more than 1300 mammogram images from both the King Hussein Cancer Center and Jordan Hospital. Results achieved tumor classification accuracy values of 90.7%. Moreover, the results showed a sensitivity of 96.2% and a specificity of 94.4% for the mass classifying algorithm. Radiologists from both institutes have acknowledged the results and confirmed that this work has lead to better visual quality images and that the segmentation and classification of tumors has aided the radiologists in making their diagnoses.

Introduction

Breast cancer originates from breast tissues, most commonly from the inner lining of milk ducts or the lobules that supply the ducts with milk, breast cancer occurs in humans and other mammals. The overwhelming majority of human cases occur in women. Mammograms are effective for breast cancer screening and early detection of masses or abnormalities. Masses and calcifications are the most common abnormalities that indicate breast cancer [30]. According to radiologists, a mass screened on a mammogram can be either benign or malignant depending on its shape. Benign tumors usually have round or oval shapes, whilst malignant tumors have a partially rounded shape with a spiked or irregular outline. Noncancerous or benign tumors include cysts, fibro adenomas, and breast hematomas. A cancerous or malignant tumor in the breast is a mass of breast tissue that grows in an abnormal and uncontrolled way. Normally, malignant masses appear brighter than any tissue surrounding it [30]. Successful diagnosis in mammography depends on detecting cancer in its earliest and most treatable stage. However, as mammogram images are subject to noise and distortion, computer aided detection (CAD) techniques are used to assist radiologists in the early detection of cancer, by enhancing those images and providing in-depth analysis of visual details in mammogram images.

Mammogram image segmentation mutually partition homogeneous regions into meaningful regions of interest. The segmentation algorithms can be categorized into two distinct categories according to the regions to be segmented, i.e., breast region segmentation and region of interest (ROI) segmentation. The former is the process of splitting the mammogram image into a breast region and a background, in order to limit the search for abnormalities on the breast region without the effect of the background on the results. The latter is the process of segmenting the suspicious regions to be analyzed for abnormalities. Furthermore, ROI segmentation can be classified into other groups, such as: segmentation using a single view and segmentation using multiple views.

Many mammogram image enhancement and segmentation techniques were proposed in literature. However, the majority of the proposed techniques share the common steps of image enhancement, segmentation, quantification, registration, and visualization. Those techniques can be differentiated by the varying algorithms employed at each step. One of the challenges faced by current mammogram image detection techniques lies in the difficulty of analyzing dense tissues. This difficulty can be attributed to the breast region which appears white in the mammogram images making masses and specifically micro-calcification highly invisible intermixed with the background tissue. Moreover, literature reveals that those approaches are designed for single view mammograms, and propose either an enhancement or a segmentation method. In this work, the proposed mammogram enhancement and segmentation algorithms are designed for mammograms taken from different views, which include: medio-lateral oblique (MLO) view and cranio-caudal view (CC).

In order to support radiologists in the detection of abnormalities, image enhancement is applied on mammogram images to increase their contrast and decrease the noise present. For this reason, this research investigates the use of different combination of conventional and feature-based contrast enhancement and noise reduction algorithms in order to enhance the visual details for radiologists and facilitate the masses segmentation process.

Moreover, in this research, a semi-automated mammogram analysis research tool is built to aid radiologists in analyzing breast regions and in detecting cancerous masses. With the help of this research, the radiologist can effectively segment the breast region and enhance its visual quality appropriately in order to clarify any abnormalities. Following the enhancement step, suspicious regions are segmented and sent to a mass classifying method to classify masses into all possible benign or malignant cases.

In addition to the introduction provided in Section 1, Section 2 provides a literature study on mammogram image enhancement and segmentation techniques, in depth information of the provided literature can be found on the authors work in [5]. Section 3 provides the proposed algorithms for enhancement and segmentation processes. Section 4 provides the experimental results and analysis. Finally Section 5 concludes this work with an insight to future work.

Section snippets

Related work

Mammogram enhancement methods are used to enhance details and reduce noise and other artifacts in X-ray images, in order to facilitate the radiologist's visual interpretation of the detection of abnormalities. Several methods are used to manipulate mammogram images, and can be categorized into four main groups; conventional enhancement techniques, region-based enhancement techniques, feature-based enhancement techniques, and fuzzy enhancement techniques. Conventional enhancing techniques appear

Proposed mammogram analysis (enhancement and segmentation) techniques

In this work, the proposed mammogram enhancement and segmentation algorithms are designed for mammograms taken from different views, which include: medio-lateral oblique (MLO) view and the cranio-caudal view (CC). The proposed mammogram analysis algorithm consists of five stages described as follows:

Stage 1: In the first stage of the proposed analysis algorithm (as shown in Fig. 1), breast region segmentation is applied and it consists of three phases: in the first phase, borders, labels, and

Experiments and results

The mammogram image dataset used in this research is collected from both the King Hussein Cancer Center (KHCC) and Jordan Hospital (JH) with a total of 1317 images. The data collected from the Jordan Hospital contained a total of 792 images as follows: 109 normal cases (cancer free), 68 benign cases, and 21 malignant cases. Each case contains two mammographic views for each breast with a resolution of 768 × 942, 452 normal mammogram images, 236 benign mammogram images that contain

Conclusion

Depending on its shape, a mass screened on a mammogram can be either benign or malignant. Usually benign tumors have round or oval shapes, while malignant tumors have a partially rounded shape with a spiked or irregular outline. Noncancerous or benign tumors include cysts, fibro adenomas, and breast hematomas. Radiologist's decisions are often challenged by the quality of mammogram images and their ability to classify different types/sizes of masses that appear in mammograms. In this research,

Acknowledgements

We would like to thank the radiology departments at Jordan Hospital (JH) and King Hussein Cancer Center (KHCC) for providing us with the required data. Special thanks to Dr. Shoman Dr. Harara, and Dr. Mereb from Jordan Hospital, Mr. Nael Al-Shishani and Ms. Lina Bassam from King Hussein Cancer Center for their assistance, evaluation, and feedback.

References (67)

  • T. Stojic et al.

    Adaptation of multifractal analysis to segmentation of microcalcifications in digital mammograms

    Phys. A Stat. Appl.

    (2006)
  • X. Sun et al.

    Ipsilateral-mammogram computer-aided detection of breast cancer

    Comput. Med. Imaging Graph.

    (2004)
  • H. Yang et al.

    LS-SVM based image segmentation using color and texture information

    J. Vis. Commun. Image Represent.

    (2012)
  • B. Zheng et al.

    Computerized detection of masses in digitized mammograms using single-image segmentation and a multilayer topographic feature analysis

    Acad. Radiol.

    (1995)
  • M. Altrichter et al.

    Joint analysis of multiple mammographic views in cad systems for breast cancer detection

  • D. Ang et al.

    Computer aided diagnosis system for digital mammogram based on neural network and dual tree complex wavelet transform

    Procedia Eng.

    (2012)
  • M. Antonie et al.

    Application of data mining techniques for medical image classification

  • M. Bajger et al.

    Automatic tuning of MST segmentation of mammograms for registration and mass detection algorithms

    Digit. Image Comput.

    (2009)
  • M. Beltawi et al.

    Mammogram enhancement and segmentation methods: classification, analysis, and evaluation

  • M. Bhattacharya et al.

    Fuzzy logic based segmentation of microcalcification in breast using digital mammograms considering multi-resolution

  • J. Blackledge

    Digital Image Processing

    (2005)
  • K. Bovis et al.

    Detection of masses in mammograms using texture features

  • A. Cao et al.

    Breast mass segmentation based on information theory

  • D.D. Cascio

    Mammogram segmentation by contour searching and mass lesions classification with neural network

    IEEE Trans. Nucl. Sci.

    (2006)
  • R. Chandrasekhar et al.

    A simple method for automatically locating the nipple on mammograms

    IEEE Transact. Med. Imaging

    (1997)
  • C. Chang et al.

    Enhancement of mammograms from oriented information

    Int. Conf. Image Process.

    (1997)
  • Z. Chen et al.

    Segmentation of the breast region with pectoral muscle removal in mammograms

  • H.D. Cheng et al.

    A novel approach to micro-calcification detection using fuzzy logic technique

    IEEE Trans. Med. Imaging

    (1998)
  • K. Chu

    An introduction to sensitivity, specificity, predictive values and likelihood ratios

    Emerg. Med. Australas.

    (1999)
  • M.L. Comer et al.

    Statistical segmentation of mammograms

  • W. Dabour

    Improved wavelet based thresholding for contrast enhancement of digital mammograms

    IEEE Control Syst. Mag.

    (2008)
  • C. D’Elia et al.

    Detection of microcalcifications clusters in mammograms through TS-MR’F segmentation and SVM-based classification

  • C. D’Elia et al.

    A tree-structured Markov random field model for Bayesian image segmentation

    IEEE Trans. Image Process.

    (2003)
  • Cited by (77)

    • Breast cancer detection model using fuzzy entropy segmentation and ensemble classification

      2023, Biomedical Signal Processing and Control
      Citation Excerpt :

      It is best to retrieve the features from the gathered information first [61]. Then it requires scaling of each dimension of the normalized data; such that they fall into the same range [20 21]. But, this turns into a time-consuming and intricate process.

    • Breast tumor segmentation in digital mammograms using spiculated regions

      2022, Biomedical Signal Processing and Control
      Citation Excerpt :

      One of the most important steps in CAD is segmentation [9]; however, this step must accurately extract the tumor shape characteristics from the mammogram [10]. To accomplish this, segmentation partitions the mass into the actual tumor and the background [11,12]. Any damage to the margin of a malignant mass during the segmentation step can cause the tumor to be recognized as being benign [13].

    View all citing articles on Scopus
    View full text