Detection of microcalcifications in digital mammograms using wavelet filter and Markov random field model

https://doi.org/10.1016/j.compmedimag.2006.03.002Get rights and content

Abstract

Clustered microcalcifcations (MCs) in digitized mammograms has been widely recognized as an early sign of breast cancer in women. This work is devoted to developing a computer-aided diagnosis (CAD) system for the detection of MCs in digital mammograms. Such a task actually involves two key issues: detection of suspicious MCs and recognition of true MCs. Accordingly, our approach is divided into two stages. At first, all suspicious MCs are preserved by thresholding a filtered mammogram via a wavelet filter according to the MPV (mean pixel value) of that image. Subsequently, Markov random field parameters based on the Derin–Elliott model are extracted from the neighborhood of every suspicious MCs as the primary texture features. The primary features combined with three auxiliary texture quantities serve as inputs to classifiers for the recognition of true MCs so as to decrease the false positive rate. Both Bayes classifier and back-propagation neural network were used for computer experiments.

The data used to test this method were 20 mammograms containing 25 areas of clustered MCs marked by radiologists. Our method can readily remove 1341 false positives out of 1356, namely, 98.9% false positives were removed. Additionally, the sensitivity (true positives rate) is 92%, with only 0.75 false positives per image. From our experiments, we conclude that, with a proper choice of classifier, the texture feature based on Markov random field parameters combined with properly designed auxiliary features extracted from the texture context of the MCs can work outstandingly in the recognition of MCs in digital mammograms.

Introduction

Breast cancer has been widely recognized as one of the most frequent, malignant tumors in women. For early detection of breast cancer, mammography is often regarded as more effective than other imaging modalities in exposing the indirect sign of malignancy—microcalcifications (MCs) in the breast tissues. Nevertheless, raw mammogram images are not apt to human perception in distinguishing normal glandular tissues and malignant disease [1], especially when the images are obtained from younger women who have denser breast tissues. This is mainly due to two problems: (1) low contrast between MCs and the surrounding tissues in the raw mammogram images, which leads to difficulty in spotting suspect MCs; and (2) high false positive rate in the final MC classification. Both problems are further aggravated by the highly variant sizes and shapes of MCs. Therefore, computer-aided analysis is crucial for digital mammograms.

Many methods have been proposed to increase the contrast of MC clusters and normal tissues in mammogram images. One school of research works focused on manifesting the MC regions in the image. For example, Gordon and Rangayyan [2] proposed a local enhancement method which took advantage of the statistical characteristics of the neighborhood of each pixel. Although this method overcomes the drawback of the global enhancement method [3], it cannot adapt to varying sizes and shapes typically found in MC clusters. To this end, Morrow et al. [4] suggested an adaptive-neighborhood technique to enhance the identification of MC clusters. However, the scheme requires both high-resolution digitization of mammograms and high-resolution processing procedures, which is often impractical in typical deployment.

The other school of research works focused on removing the background noise so as to increase the contrast. For example, Lai et al. [5] investigated a selective median filtering, in which the tradeoff between noise removal and edge preservation is embedded in the decision of the filtering threshold. Zhao [6] employed morphology to remove the background noise. But this method requires a prior knowledge of the size and shape of MC clusters, which is often not available during initial identification. Zhao et al. [7] also employed an adaptive threshold function from morphological operations in conjunction with a rule base provided by expert radiologist to identify suspected MC clusters for further clinical diagnosis. Strickland and Hahn [8] developed a two-stage method based on wavelet transform for detecting and segmenting MC clusters. They demonstrated that with proper wavelet basis, the detection of MCs can be easily achieved and a straightforward thresholding can be applied to segment them. Following [8], Wang and Karayiannis [9] presented an MC detection method that consisted of decomposing the mammogram into different frequency subbands, suppressing the low-frequency subband, and finally, reconstructing the mammogram from the subbands containing only high frequencies.

Meanwhile, research works have also been conducted to tackle the high false rate problem in final MC classification. For example, Zheng et al. [10] proposed a mixed-feature-based neural network (MFNN) using spectral entropy as the decision criterion to detect MC clusters. Yu and Guan [11] extracted 31 features (mean, variance, contrast, shape, etc.) and employed a general regressive neural network (GRNN) to detect clustered MCs. Gulsrud and Husoy [12] used a filter optimized with respect to the Fisher criterion to extract texture features for the distinction between MCs and normal tissues. El-Naqa et al. [13] employed support vector machines (SVM) to detect MCs and proposed a successive enhancement learning scheme to reduce the false positives.

This paper proposes a two-stage computer-aided image analysis method to overcome the two main problems in the detection of MCs in mammograms, namely the low contrast between the MCs and the surrounding breast tissue and the high false positive rate in MCs classification. In the first stage, a wavelet filter is utilized to remove the background noise and preserve all the suspicious MCs in the images. In the second stage, Markov random field parameters are used to enhance the recognition of ‘true’ MCs and thus reduce the false positives rate. A group of 20 mammograms from the MiniMammographic Database provided by the Mammographic Image Analysis Society [14] is used to verify the effectiveness of the proposed method.

The rest of this paper is organized as follows: Section 2 presents a brief description of mathematical foundation of wavelet analysis and Markov random field model. Section 3 describes the method of suspicious MCs detection, feature extraction, and classification. Section 4 presents our experimental results and the related discussion. Section 5 is the conclusion remarks of this study.

Section snippets

Multiresolution representation of signal with wavelets

The microcalcifications (MCs) in mammograms usually appear in small clusters of a few pixels with relatively high intensity. In this regard, wavelet transform presents an excellent analysis tool since it uses shorter windows at high frequencies and long windows at low frequencies, which effectively separates the high frequency MCs from the low frequency background tissues in the mammograms. In the following, we briefly review the theory of wavelet representation needed in this paper [15], [16],

The proposed method

As we mentioned in the context, recognition of microcalcifications (MCs) in digital mammograms involves two critical issues: (1) detection of suspicious MCs and (2) recognition of true MCs. We propose a two-stage strategy to treat these issues separately. In the first stage, a wavelet filter is exploited to detect all suspicious MCs. In the second stage, texture features based on a Markov random field model are extracted and serve as inputs to the MC classifiers. Fig. 6 shows the block diagram

Experimental results and discussions

The MiniMammographic Database [2] provided by the Mammographic Image Analysis Society (MIAS) was used as the testbench. The database contains mammogram images acquired from 322 different women, which were divided into seven categories according to different pathological types. All mammograms are digitized at 200-μm pixel edge, giving a 1024×1024 pixel resolution for each mammogram. The microcalcification (MC) category in the database contains 20 images with MC numbers ranging from 1 to 3,

Conclusion

In this paper, we proposed a two-stage method to detect microcalcifications (MCs) in digital mammograms. In the first stage, a wavelet filter was used to identify suspicious MCs from the mammograms. In the second stage, features based on a Markov random field (MRF) model and three auxiliary texture quantities were extracted from the neighborhood of the suspicious MCs and classified by a Bayes classifier and a BPNN. Twenty mammograms containing 25 areas of MCs from the MIAS MiniMammogram

Acknowledgements

This work was supported in part by the grant NSC90-2213-E-194-037 from the National Science Council, Taiwan. The authors would like to thank Prof. Ching-Wei Yeh in the EE Department of NCCU for many valuable discussions and comments.

Sung-Nien Yu obtained his BS and MS degrees in Electrical Engineering from National Taiwan University, Taipei, Taiwan in 1987 and 1991, respectively. He obtained his PhD degree in Biomedical Engineering from Case Western Reserve University in 1996. He is currently an Assistant Professor in the Department of Electrical Engineering at National Chung Cheng University, Chia-Yi, Taiwan. His research interests include digital signal processing, image processing, and the application of information

References (30)

  • H. Derin et al.

    Segmentation of textured images using Gibbs random fields

    CVGIP Graph Mod Image Process

    (1986)
  • C.O. Acuna

    Texture modeling using Gibbs distribution

    CVGIP Graph Mod Image Process

    (1992)
  • P.C. John et al.

    X-ray characterization of normal and neoplastic breast tissues

    Phys Med Biol

    (1987)
  • R. Gordon et al.

    Feature enhancement of film mammograms using fixed and adaptive neighborhoods

    Appl Opt

    (1984)
  • A.K. Jain

    Fundamentals of digital image processing

    (1989)
  • W. Morrow et al.

    Region-based contrast enhancement of mammograms

    IEEE Trans Med Imaging

    (1992)
  • S. Lai et al.

    On techniques for detecting circumscribed masses in mammograms

    IEEE Trans Med Imaging

    (1989)
  • Zhao D, Shridhar M, Daut DG. Morphology on detection of calcifications in mammograms. Proceedings of IEEE int. conf...
  • D. Zhao

    Rule-based morphological feature extraction of microcalcifications in mammograms

    SPIE Med Imaging

    (1993)
  • R.N. Strickland et al.

    Wavelet transforms for detecting microcalcifications in mammograms

    IEEE Trans Med Imaging

    (1996)
  • T.C. Wang et al.

    Detection of microcalcifications in digital mammograms using wavelets

    IEEE Trans Med Imaging

    (1998)
  • B. Zheng et al.

    Digital mammography: mixed feature neural network with spectral entropy decision for decision of microcalcifications

    IEEE Trans Med Imaging

    (1996)
  • S. Yu et al.

    A CAD system for the automatic detection of clustered microcalcifications in digitized mammogram films

    IEEE Trans Med Imaging

    (2000)
  • T.O. Gulsrud et al.

    Optimal filter-based detection of microcalcifications

    IEEE Trans Biomed Eng

    (2001)
  • I. El-Naqa et al.

    A support vector machine approach for detection of microcalcifications

    IEEE Trans Med Imaging

    (2002)
  • Cited by (69)

    • Microcalcification cluster detection using multiscale products based Hessian matrix via the Tsallis thresholding scheme

      2017, Pattern Recognition Letters
      Citation Excerpt :

      Set of features are extracted from image and classifying Mcs using backpropagation neural network. Sung-Nien et al. [16] used Bayes classifier and backpropagation neural network to achieve the detection process. In their method, features based on Markov random field parameters were taken.

    • A new method of micro-calcifications detection in digitized mammograms based on improved simplified PCNN

      2016, Neurocomputing
      Citation Excerpt :

      The area under the ROC curve (AUC) can be thought as the probability that the test result from a randomly chosen abnormal case is more indicative of abnormal than that from a randomly chosen normal individual [22]. Our method almost has the same performance as Ref. [9]. The Youden's index of the proposed method is up to 0.91362.

    • Local energy-based shape histogram feature extraction technique for breast cancer diagnosis

      2015, Expert Systems with Applications
      Citation Excerpt :

      The literature reports many techniques for feature extraction. Yu, Li, and Huang (2006) used a combination of model-based and statistical texture features to detect microcalcifications. This approach first detects the region of interest (ROI)—in this case, the area that contains the microcalcifications—by using wavelets and thresholds.

    • Computerized Detection of Lesions in Diagnostic Images with Early Deep Learning Models

      2022, Machine and Deep Learning in Oncology, Medical Physics and Radiology, Second Edition
    • Hybrid mechanism for medical data classification

      2021, International Journal of Advanced Intelligence Paradigms
    • Deep learning models for medical image analysis: challenges and future directions

      2019, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
    View all citing articles on Scopus

    Sung-Nien Yu obtained his BS and MS degrees in Electrical Engineering from National Taiwan University, Taipei, Taiwan in 1987 and 1991, respectively. He obtained his PhD degree in Biomedical Engineering from Case Western Reserve University in 1996. He is currently an Assistant Professor in the Department of Electrical Engineering at National Chung Cheng University, Chia-Yi, Taiwan. His research interests include digital signal processing, image processing, and the application of information technologies to biomedical problems.

    Kuan-Yuei Li obtained his BS degree in Electrical Engineering from National Cheng-Kung University, Tainan, Taiwan, ROC, in 2002. After 2 years, he obtained his MS degree in EE Department from National Chung-Cheng University, Chia-Yi, Taiwan, ROC, in 2004. He is currently a software engineer in APPRO Technology, Inc., Taipei, Taiwan. His research interests include image processing and embedded system designing.

    Yu-Kun Huang received his BS degree in civil engineering and MS degree in applied mechanics from National Taiwan University, Taipei, Taiwan, ROC, in 1987 and 1989, respectively. He is currently an Associate Professor in the Department of Civil Engineering, Nan Joen Institite of Technology, Tainan, Taiwan. Presently, he is pursuing his PhD in the Department of Electrical Engineering, National Chung Cheng University, Taiwan, ROC. His research interests span analysis and design of structures, dynamical systems, and image analysis.

    View full text