Share Email Print

Spie Press Book

Recent Advances in Breast Imaging, Mammography, and Computer-Aided Diagnosis of Breast Cancer
Editor(s): Jasjit Singh Suri; Rangaraj M. Rangayyan
Format Member Price Non-Member Price

Book Description

Breast cancer is the most common type of cancer found in women worldwide; approximately 10% of women are confronted with breast cancer in their lives. Breast cancer can be most efficiently treated if detected at an early stage. This book focuses primarily on the application of computer vision for early lesion identification in mammograms and breast-imaging volumes through computer-aided diagnostics (CAD). Color illustrations are included in the text, and an accompanying CD-ROM contains other full-color images.

The book is divided into four parts. In Part I, the anatomic, histopathologic, and mammographic views of the breast are examined, and the physics for different breast-imaging modalities are presented. Part II presents techniques for lesion and mass detection, and computer-aided diagnosis. Part III discusses the applications of different computer-vision fields in breast-image registration, and performance evaluation for breast CAD techniques is discussed in Part IV.


Book Details

Date Published: 10 April 2006
Pages: 1008
ISBN: 9780819460813
Volume: PM155
Errata

Table of Contents
SHOW Table of Contents | HIDE Table of Contents
Preface / xv
Chapter 1. The Subgross Morphology of Normal and Pathologically Altered Breast Tissue / 1
Tibor Tot
1.1 Introduction / 2
1.2 Subgross Breast Anatomy / 3
1.3 The Normal Mammogram / 16
1.4 In Situ Development of Malignant Tumors / 29
1.5 Invasive Carcinomas / 36
1.6 Time as Morphologic Factor / 50
References / 53
Chapter 2. Fundamentals of Achieving Functional and Anatomical Imaging with Breast MRI: Physical and Clinical Concepts / 57
Dee H. Wu, Ann G., Archer, L. Jill Hast, Barry J. Greer, Brenda Elledge, Heather R. Webb, Douglas Beall, Max Walter, Susan M. Edwards, and Erika Rubesova
2.1 Introduction / 58
2.2 Consequences of Composition, Architectural, and Functional Evaluation in Breast MR Imaging / 64
2.3 Breast Composition and Hormonal Influences (Fat, Ducts, Secretions, and Breast Implants) / 65
2.4 Region of Interest Definition of Tumors / 68
2.5 Technological Fundamentals of Breast MRI / 70
2.6 Summary / 108
Acknowledgments / 108
References / 108
Chapter 3. An Indexed Atlas of Digital Mammograms for Computer-Aided Diagnosis of Breast Cancer / 115
Hilary Alto, Rangaraj M. Rangayyan, Raman B. Paranjape, J.E. Leo Desautels, and Heather Bryant
3.1 Introduction / 116
3.2 Databases of Mammograms / 117
3.3 The Indexed Atlas Concept / 118
3.4 Design of the University of Calgary Indexed Atlas of Mammograms / 120
3.5 Use of Mobile Software Agents with the Indexed Atlas / 125
3.6 Discussion / 126
Acknowledgments / 128
References / 128
Chapter 4. Genetic Algorithms in CAD Mammography / 135
Renato Campanini and Nico Lanconelli
4.1 Introduction / 136
4.2 CAD Critical Issues / 139
4.3 Genetic Algorithms: A Brief Overview / 142
4.4 GA for Parameters Optimization / 147
4.5 GA for Feature Selection / 157
References / 162
Chapter 5. Mammographic Mass Detection by Robust Learning Algorithms / 165
A.Z. Cao, Q. Song, X.L. Yang, Z.M. Wang, and Y. Shui
5.1 Introduction / 166
5.2 Robust Information Clustering for Automatic Mass Detection / 167
5.3 Vicinal Support Vector Machine for Mass-Pattern Analysis / 184
References / 198
Chapter 6. Analysis of Bilateral Asymmetry in Mammograms via Directional Filtering with Gabor Wavelets / 203
Ricardo J. Ferrari, Rangaraj M. Rangayyan, J.E. Leo Desautels, and Annie F. Fr�re
6.1 Introduction / 204
6.2 Bilateral Asymmetry in Mammograms / 206
6.3 Construction of Gabor Wavelets / 206
6.4 Directional Analysis / 211
6.5 Analysis of Asymmetry in Mammograms / 217
6.6 Results / 221
6.7 Discussion and Conclusions / 230
Acknowledgments / 230
References / 230
Chapter 7. Support Vector Machines in CAD Mammography / 235
Renato Campanini, Nico Lanconelli, and Matteo Roffilli
7.1 Introduction / 236
7.2 Support Vector Machines: A Qualitative Approach / 237
7.3 SVMs for CAD / 243
7.4 Novel Featureless Approach / 251
7.5 SVMs: Mathematical Details / 258
References / 266
Chapter 8. A Weighted Gaussian Mixture Model with Markov Random Fields and Adaptive Expert Combination Strategy for Segmenting Masses in Mammograms / 269
Sameer Singh and Keir Bovis
8.1 Introduction / 270
8.2 Weighted Gaussian Mixture Models / 272
8.3 Combination of Experts / 275
8.4 Experimental Details / 280
8.5 Results / 283
8.6 Conclusions / 292
References / 292
Chapter 9. Detection of Microcalcifications in Mammograms / 297
Bego�a Acha, Carmen Serrano, Rangaraj M. Rangayyan, and J.E. Leo Desautels
9.1 Detection of Microcalcifications: Background and Motivation / 298
9.2 State of the Art / 299
9.3 Detection of Microcalcifications Using the 2D Linear Prediction Error / 303
9.4 Results / 311
9.5 Conclusions / 312
References / 314
Chapter 10. Using Computational Intelligence for Computer-Aided Diagnosis of Screen Film Mammograms / 321
Walker H. Land, Jr., Daniel W. McKee, Frances R. Anderson, Timothy Masters, Joseph Y. Lo, Mark Embrechts, and John Heine
10.1 Introduction / 323
10.2 Evolutionary Programming and Evolutionary Computation / 323
10.3 Evolutionary Programming/Adaptive Boosting Hybrid / 332
10.4 Another Evolutionary Programming Approach / 339
10.5 EP-Derived Support Vector Machines / 345
10.6 A Support Vector Machine (SVM) Generalized Regression Neural Network (GRNN) Oracle / 356
10.7 Partial Least Squares (PLS) and Kernel-Partial Least Squares
(K-PLS) / 364
10.8 Artificial Intelligence, Knowledge Representation, and Knowledge Engineering / 373
Acknowledgments / 378
References / 378
Chapter 11. New Approach for Breast Skin-Line Estimation in Mammograms / 383
Yajie Sun, Jasjit S. Suri, Rangaraj M. Rangayyan, and J.E. Leo Desautels
11.1 Introduction / 384
11.2 A Brief Overview of the Proposed System / 386
11.3 Initial Skin-Line Boundary Estimation / 388
11.4 Initial Confirmed Portion Extraction: A Greedy Approach / 390
11.5 Extraction of Stroma Edge / 393
11.6 Skin-line Estimation: A Dependency Approach / 397
11.7 Performance Evaluation System / 401
11.8 Results: Dependency Approach versus Deformable Model / 406
11.9 Discussion and Conclusions / 410
Acknowledgments / 411
References / 411
Chapter 12. Computerized Mass Detection for Digital Breast Tomosynthesis / 415
Robert Nishikawa and Ingrid Reiser
12.1 Introduction / 416
12.2 CADe Based on the Reconstructed Breast Volume / 417
12.3 CADe Based on the Sequence of Projection Images / 427
12.4 Summary / 431
Acknowledgments / 432
References / 432
Chapter 13. Breast Imaging Registration: A Review / 435
Yujun Guo, Jasjit S. Suri, and Radhika Sivaramakrishna
13.1 Introduction / 436
13.2 Registration Background / 440
13.3 Intramodality Breast-Image Registration / 446
13.4 Intermodality Breast-Image Registration / 464
13.5 Biomechanical Breast Model / 470
13.6 Validation of Breast-Registration Methods / 473
13.7 Discussion and Conclusions / 477
References / 480
Chapter 14. A Mammographic Registration Framework Based on Anatomical Linear Structures / 493
Robert Mart�, Caroline Rubin, Erika R.E. Denton, and Reyer Zwiggelaar
14.1 Introduction / 495
14.2 Correspondence of Mammographic Images / 495
14.3 Identifying Salient Linear Structures / 508
14.4 Registration Using Linear Structures / 515
14.5 Detection of Abnormalities / 520
14.6 Tracking of Linear Structures / 522
14.7 Automatic MR Correspondence Based on Linear Structures / 525
14.8 Conclusions / 531
References / 532
Chapter 15. AMDI - Indexed Atlas of Digital Mammograms that Integrates Case Studies, E-Learning, and Research Systems via the Web / 535
Denise Guliato, Ernani V. de Melo, Ricardo S. B�aventura, and Rangaraj M. Rangayyan
15.1 Breast Cancer and Mammography / 536
15.2 Indexed Atlases and Telemedicine / 536
15.3 Overview of AMDI / 537
15.4 Conclusion / 557
Acknowledgments / 559
References / 559
Chapter 16. Detection, Characterization, and Visualization of Breast Cancer Using 3D Ultrasound Images / 563
Daisuke Fukuoka, Takeshi Hara, and Hiroshi Fujita
16.1 Introduction / 564
16.2 Material and Methods / 564
16.3 Results / 569
16.4 Conclusions / 569
References / 570
Chapter 17. Computer-Aided Diagnosis for 2D/3D Breast Ultrasound / 573
Ruey-Feng Chang, Chia-Ling Tsai, Sheng-Fang Huang, Wen-Jia Kuo, Dar-Ren Chen, Woo Kyung Moon, Yu-Len Huang, Wei-Ming Chen, and Wen-Jie Wu
17.1 Introduction / 575
17.2 Texture-Based Computer-Aided Diagnosis for 2D Breast Ultrasound / 576
17.3 Shape-Based Computer-Aided Diagnosis for 2D Breast Ultrasound / 581
17.4 Classification Techniques for Breast US CAD / 586
17.5 Computer-Aided Diagnosis for 3D Breast Ultrasound / 594
17.6 Conclusion / 611
References / 612
Chapter 18. Automatic Pectoral Muscle Segmentation on Mediolateral Oblique View Mammograms / 617
Sze Man Kwok, Ramachandran Chandrasekhar, Yianni Attikiouzel, and Mary T. Rickard
18.1 Introduction / 618
18.2 Overall Algorithm / 620
18.3 Orientation and Notation / 621
18.4 Straight-Line Estimation and Validation / 622
18.5 Iterative Cliff Detection / 627
18.6 Experimental Setup / 631
18.7 Results / 632
18.8 Discussion / 636
18.9 Conclusions / 639
Acknowledgments / 640
References / 640
Chapter 19. Performance evaluation of CADe algorithms in mammography / 643
M.A. Wirth
19.1 Introduction / 645
19.2 Performance Evaluation in CADe / 649
19.3 Evaluating Image-Processing Algorithms / 654
19.4 Evaluating Enhancement Algorithms / 660
19.5 Little Boxes Made of Algorithms / 664
19.6 Performance Evaluation Framework / 665
19.7 Mammogram Databases / 671
19.8 Breast-Region Segmentation: Examination of a Performance-Evaluation Process / 676
19.9 Recommendations / 688
References / 691
Chapter 20. A Simple Method for Automatically Locating the Nipple on Mammograms / 705
Ramachandran Chandrasekhar and Yianni Attikiouzel
20.1 Introduction / 706
20.2 Review of Previous Work / 706
20.3 Outline of Method / 707
20.4 Notation / 708
20.5 Detailed Description of Algorithm / 709
20.6 Experimental Method / 712
20.7 Results and Discussion / 715
20.8 Conclusions / 724
Acknowledgments / 724
References / 725
Chapter 21. Algorithms for Segmenting Small Low-Contrast Objects in Images / 727
Isaac N. Bankman
21.1 Introduction / 728
21.2 Segmentation Algorithms / 731
21.3 Results / 735
21.4 Conclusion / 740
References / 741
Chapter 22. Computer-Aided Diagnosis of Breast Cancer on MR Imaging / 743
Qiu Wu and Mia K. Markey
22.1 Medical Imaging for Breast Cancer Detection and Diagnosis / 744
22.2 Magnetic Resonance Imaging of the Breast / 744
22.3 Dynamic Contrast-Enhanced Breast MRI / 745
22.4 Computer-Aided Detection and Diagnosis / 748
22.5 Developing CADe/CADx for DCE Breast MRI / 749
22.6 Future Directions / 757
Acknowledgments / 759
References / 759
Chapter 23. Review on Real-Time Magnetic Resonance Gad-Enhanced Breast Lesion Characterization / 767
Jasjit S. Suri, A.X. Falcao, L. Reden, J. Gao, and S. Laxminarayan
23.1 Introduction / 768
23.2 Data Acquisition and Scanning Protocols / 770
23.3 Three Techniques for MR Breast Tumor Detection and Analysis / 773
23.4 Issues Related to Real Time and Graphical User Interfaces / 786
23.5 Conclusions and Future Challenges / 788
Acknowledgments / 788
References / 789
Chapter 24. Breast MRI Computer-Aided Diagnosis Systems / 795
Lina Arbash Meinel and Joseph M. Reinhardt
24.1 Introduction / 796
24.2 Breast MRI Segmentation / 797
24.3 Selection of the Shape, Texture, and Contrast-Enhancement Features / 817
24.4 Breast MRI Classification / 823
24.5 Validation / 830
24.6 Future research / 831
References / 831
Chapter 25. Automatic Assessment of Mammogram Adequacy and Quality / 837
Ramachandran Chandrasekhar and Sze Man Simon Kwok
25.1 Introduction / 838
25.2 The Mediolateral Oblique View Mammogram / 840
25.3 Adequacy Assessment Algorithms / 841
25.4 Results / 847
25.5 Discussion / 847
25.6 Conclusions / 848
Acknowledgments / 848
References / 848
Chapter 26. Probability Modeling of CAD Systems for Mammography / 851
John Maleyeff and Frank C. Kaminsky
26.1 Introduction / 852
26.2 Background and Motivation / 854
26.3 Probability Modeling in CAD / 855
26.4 Analysis of Mammography Systems / 861
26.5 Illustrative Model Results / 865
26.6 Conclusion / 870
References / 872
Chapter 27. Computer-Aided Diagnosis in Breast Imaging: Where Do We Go after Detection? / 875
Joseph Y. Lo, Anna O. Bilska-Wolak, Jay A. Baker, Georgia D. Tourassi, Mia K. Markey, and Carey E. Floyd
27.1 Introduction / 876
27.2 CADx Classifier Models / 877
27.3 Feature Selection / 883
27.4 Data Sampling / 886
27.5 Clinical and Regulatory Acceptance / 890
References / 896
Chapter 28. The Current Status and Likely Future of Breast Imaging CAD / 905
Jasjit S. Suri, Ramachandran Chandrasekhar, Nico Lanconelli, Renato Campanini, Matteo Roffilli, Ruey-Feng Chang, Yujun Guo, Radhika Sivaramakrishna, Tibor Tot, Bego�a Acha, Carmen Serrano, Ingrid Reiser, Robert M. Nishikawa, Dee H. Wu, Koon-Pong Wong, Ashwini Kshirsagar, Yajie Sun, Michael Wirth, Aize Cao, J.E. Leo Desautels, and Rangaraj M. Rangayyan
28.1 Introduction / 907
28.2 Shift in Breast Imaging Paradigms / 909
28.3 Different Imaging Modalities for Breast Imaging / 909
28.4 Current and Future Challenges for CAD / 922
28.5 Commercial Breast CAD Systems and Their Role in Clinical Settings / 939
28.6 Resources in Breast Imaging: Research Centers, Conferences, Databases, Publications, and Patents / 944
28.7 Conclusions / 946
Acknowledgment / 948
References / 948
Index / 967

Preface

Breast cancer is the most common type of cancer in women worldwide. About ten percent of women are confronted with breast cancer in their lives. Breast cancer can be most efficiently treated if detected at an early stage. This book focuses on the application of computer vision for lesion identification in mammograms and breast imaging volumes called computer-aided diagnosis (CAD).

The book is divided into four parts: Part I presents the anatomic, histopathology, and mammographic views of the breasts, and the physics of different breast imaging modalities. Part II presents the techniques for lesion detection, mass detection through CAD. Part III presents the applications of different computer vision fields in breast-image registration. Finally, Part IV presents the performance evaluation section for breastCAD techniques.

Chapter 1 summarizes 20 years of Dr. Tibor Tot�s experience in routine diagnostics using breast imaging. The group at Uppsala University, the Sweden has more than 5,000 operated breast lesions documented on large 2D histological sections, more than 400 of them are also examined in thick 3D sections. Their statistical work-up was carried out during the period of 1996�2003, corresponding to more than 1,700 breast carcinoma cases. This chapter not only aims to describe the morphologic findings in their static state, but attempts to place them in a perspective of the dynamic changes involving pathologically altered breast tissue: an ambition similar to describing a life-long movie with a couple of words and photographs.

Breast MRI has shown great potential as a diagnostic tool and is rapidly becoming part of the standard of care in breast cancer evaluation. The Chapter 2 describes the clinical science of breast MRI from clinical presentation, composition, and scientific first principles. Major technological advances and recent emergent technologies in breast MRI, including anatomically oriented pulse sequences, human factors in imaging, and molecular imaging techniques are discussed. The reader gently introduced to concepts, clinical presentations, and manifestations of disease. As such, the goal is to provide a resource for scientists and clinicians to be able to better understand the multitude of decision-making capabilities of breast MRI by understanding established techniques, as well as emerging new technologies for this major women�s health issue.

Computer-aided diagnostic systems are being developed to assist radiologists in the interpretation of ambiguous mammographic features corresponding to possible signs of early breast cancer. Databases of digital mammograms are needed for testing such systems. Chapter 3 presents an overview of a few such databases. Most databases are limited to single-exam sets of two or four mammograms on which a diagnosis was made, some ground-truth information related to the position of diagnostically significant mammographic features, and the diagnosis. The chapter presents the design of a comprehensive, indexed atlas of digital mammograms. The design of an appropriate indexing scheme facilitates the implementation of content-based retrieval techniques needed for efficient access to and retrieval of relevant cases from the atlas. Finally, the chapter proposes the use of mobile software agents for facilitating remote consultation of the atlas. Mobile agents can move between data sources such as the atlas and hospital repositories, perform computational tasks at each site, and return only relevant data to the user. These features reduce the computational requirements of the local computer system, bandwidth requirements, and overall network traffic. Proposed applications of the atlas include research, remote consultation, teaching, evaluation of CAD systems, and self-evaluation by radiologists.

The Chapter 4 focuses on the application of genetic algorithms (GAs) in computer-aided diagnostics (CAD). An overview of the GAs is supplied, with a special focus on their use as optimization techniques. In particular, the use of GAs for optimizing the choice of the numerous parameters involved in a CAD detection method is described. Major advantages induced by the use of a GA in CAD detection issues are described, both in term of performance improvement and of the investigation of the importance of the different CAD parameters.

Effective mass detection and analysis are important steps in mammography. Chapter 5 discusses a study showing two aspects: (a) automatic detection of masses surrounded by glandular or dense glandular breast tissues, based on robust image segmentation techniques clustering information to ensure that suspicious masses are isolated from background breast tissue and in the number of False Positive (FP) are as few possible; and (b) mass pattern analysis to classify the region of interest (ROI) into normal or abnormal, and benign or malignant cases by a semi-automatic analysis approach, where a new classification of vicinal support-vector machine is proposed. Experimental results based on the Mini-MIAS* database demonstrate that the proposed algorithms are effective for mammographic mass detection and analysis.

Chapter 6 presents a procedure for analysis of left�right (bilateral) asymmetry in mammograms. The procedure is based on the detection of linear-directional components using a multiresolution representation based upon Gabor wavelets. A particular wavelet scheme with 2D Gabor filters as elementary functions with a varying tuning frequency and orientation, specifically designed in order to reduce the redundancy in the wavelet-based representation, is applied to the given image. The filter responses for different scales and orientation are analyzed by using the Karhunen� Lo�ve (KL) transform and Otsu�s method of thresholding. The KL transform is applied to select the principal components of the filter responses, preserving only the most relevant directional elements appearing at all scales. The selected principal components, thresholded by using Otsu�s method, are used to obtain the magnitude and phase of the directional components of the image. Rose diagrams computed from the phase images and the statistical measures computed thereof are used for quantitative and qualitative analysis of the oriented patterns. A total of 80 images from 20 normal cases, 14 asymmetric cases, and six architectural distortion cases from the Mini- MIAS database were used to evaluate the scheme using the leave-one-out methodology. Average classification accuracy rates of up to 74.4% were achieved.

Chapter 7 focuses on the use of machine learning techniques such as support vector machines (SVM) in CAD issues. First, an introduction of SVMs is presented, with a particular attention to their use as classifiers. A brief theory preamble and a survey of the advantages of SVMs over other classifiers is provided. Finally, examples of mass and microcalcification detection based on SVMs are described and reviewed. Particular emphasis is given to detection techniques that do not make use of extracted features for isolating the suspect regions.

In Chapter 8 a novel methodology for the unsupervised and supervised segmentation of mammograms for the purpose of detecting masses using weighted Gaussian mixture models andMarkov random fields (MRF) is proposed consisting of a total of four image segmentation models. The results of implementing these models on the randomly selected Brodatz composite images and DDSM database are shown and discussed. Then a novel adaptive weighted model for the based on different features and, combined of experts using one of the four segmentation models, show it to perform significantly better at 5% significance level to the conventional ensemble-combination rules strategy.

Chapter 9 first presents a small review of most significant existing methods to detect microcalcifications in mammograms. Next, a method developed by the authors for detecting microcalcification is exposed. The method is based on a seed selection procedure, which consists in an analysis of the prediction error obtained after applying a 2D linear-prediction-adaptive filter to the image. Subsequently, a region growing algorithm is performed in order to identify the microcalcifications. The method was tested on 23 mammograms from the Florida DDSM database. Two parameters were calculated to measure the performance of the algorithm: sensitivity (S) and positive predictive value (PPV), obtaining a value of S = 82.72% and PPV = 88.39%, respectively.

Chapter 10 discusses the application of several computational intelligence paradigms to the classification and diagnosis of breast cancer using mammogram screen-film data sets from several different institutions. Specifically the theory, application, and results obtained for the following paradigms has been discussed: evolutionary programming, evolutionary programming/adaptive boosting hybrid, probabilistic neural network (PNN), generalized regression neural network (GRNN), support vector machines (SVM) with several different kernels, SVM/GRNN oracle hybrid, differential evolution, partial least squares, and kernel partial least squares. The approach discusses the theories of each of these paradigms followed by a description of their specific application to screen-film mammogram data sets. These discussions are followed by evaluating these theories against specific measures of performance such as area under the receiver operating characteristic (ROC) curve, partial Az, specificity, and positive predictive values (PPV) and specificities at clinical relevant sensitivities.

Accurate breast skin-line estimation is an important prerequisite for both enhancement and analysis of mammograms for computer-aided diagnosis of breast cancer. In Chapter 11, the authors proposed a novel system for skin-line estimation. First an initial estimate of the skin line is first computed using a combination of adaptive thresholding and connected-component analysis. Due to noise, this skin line is susceptible to errors in the top and bottom portion of the breast region. Using the assumption, that the Euclidean distance from the edge of the stroma to the actual skin line is usually uniform, the chapter develops a novel dependency approach for estimating the skin-line boundary of the breast. In this dependency approach, first the constraints are developed between the stroma edge and initial skin-line boundary using the Euclidean distance. These constraints are then propagated to estimate the upper and lower skin- line portions. The selection of the constrained region is based on a greedy algorithm, which is also a new component in the system. The authors evaluate the performance of their skin-line estimation algorithm by comparing the estimated boundary with respect to the ground-truth boundary drawn by an expert radiologist. Two different metrics for error measurement are used: the polyline distance measure, and the Hausdorff distance measure. As part of our protocol, the dependency approach methodology is compared with a deformable model strategy (proposed by Ferrari et al. [1]). On a dataset of 82 images from the MIAS database, the dependency approach yielded a mean error (�) of 3.28 pixels with a standard deviation (?) of 2.17 pixels using the polyline distance measure. In comparison, the deformable model strategy yielded � = 4.92 pixels with ? = 1.91 pixels. The improvement is statistically significant. The results are clinically relevant and appealing according to the radiologists who evaluated the results.

Chapter 12 is the application of CAD on 3D digital tomosynthesis data sets. Digital tomosynthesis has recently emerged as a new and promising modality in breast imaging. Digital breast tomosynthesis (DBT) produces a quasi-3D data set that is reconstructed from a small number of projections of the compressed breast. The 3D data set consists of 1-mm thick slices through the breast. As a result, the radiologist is typically required to read 50 images of each breast. Computer-aided detection (CADe) may be useful in helping radiologists read this large image set. The chapter shows two different approaches to detect mass lesions in the DBT. The first is a 3D method to examine the reconstructed image slices. The second is to operate directly on the set of raw projection images. The second technique has several advantages including independence from the exact reconstruction algorithm used. This is important since reconstruction algorithms for this DBT are not yet fully optimized and can produce considerable artifacts in the reconstructed slices.

It is known that a number of breast imaging methods for diagnosis and biopsy of suspicious lesions are available. X-ray mammography, magnet resonance imaging (MRI), and sonography are the mostly used methods. Chapter 13 focuses on image registration as an important problem in breast imaging. It is used in a wide variety of applications that include better visualization of lesions on pre- and postcontrast breast MRI images, speckle tracking and image compounding in breast ultrasound images, alignment of positron emission, and standard mammography images on hybrid machines, etc. It is a prerequisite to align images taken at different times, using different imaging modalities, on different breast sections. The inhomogeneous, anisotropic nature of the soft-tissue within the breast, and its inherent nonrigid body behavior, in addition to temporal changes of the breast tissue, breast position, and the imaging conditions, all make breast-image registration a challenging task. This chapter gives an overview of the current state-of-the-art in the breast-image registration techniques. Methods are classified according to the modalities involved in the registration process. For the intramodality registration techniques, x-ray and MRI are the primary focus of interest in the literature. Multimodality techniques cover the combination of x-ray and ultrasound, x-ray and MRI, and more.

Two-dimensional mammography is the most common modality and 3D MRI is second best. The Chapter 14 focuses on multilevel framework for 2D and 3D registration of mammographic data, which uses the following distinct steps: (a) extraction of anatomical linear structures, (b) selection of salient linear structures, (c) extraction of local features, and (d) matching the set of feature points. Temporal and contralateral 2D x-ray and temporal 3D MR registration are used as examples.

Chapter 15 describes AMDI � Atlas Indexado de Mamografias Digitais, an indexed atlas of digital mammographic images that includes four views for each case, the patient�s clinical history, global and specific radiological findings, contours of the breast and diagnostic features, and details of features, such as the number of individual calcifications present in each cluster as identified by an authorized specialist. AMDI integrates three systems: a mammogram database registration system, a research system, and a teaching system. AMDI includes concepts related to Web-based medical databases and ontologies for the representation of knowledge in the area of mammography and computer-aided diagnosis of breast cancer. The system facilitates the entry of new cases and related mammograms, radiological information, clinical information, and information related to the health and lifestyle of a patient by a registered and authorized clinical user. The atlas incorporates a teaching system and a research system. The systems are designed to permit computer-aided diagnosis, teleradiology, telemedicine, content-based case or image retrieval, data mining, and distance learning. All the resources of AMDI are available via the Web to authorized users.

Chapter 16 presents the current status of breast screening in Japan and the sophisticated detection, classification, and visualization techniques are described. Breast screening using ultrasound images has started in some regions of Japan for early detection of breast cancer. Although the mammography is also recommended for the screening, the effectiveness of ultrasound testing is recognized for the dense breast of women aged over 40. A CAD scheme for detection of breast masses using free-hand probe and 3D imaging ultrasound devices has been developed. The scheme includes a detection technique of breast lesions based on active contour and balloon models in 2D and 3D spaces, a classification step using statistical approaches using image features, and a fine visualization browser using virtual Band C-mode images to confirm the shape and the position of the lesion.

Ultrasound is a valuable adjunct to mammography in breast imaging, in addition to distinguishing cysts from solid breast tumors. Chapter 17 focuses on 3D ultrasound for distinguishing benign and malignant lesions. Ultrasound can be used to help differentiate benign from malignant solid masses. Ultrasound criteria for the classification of solid breast masses include lesion shape, orientation, margin, echogenicity, and acoustic transmission. Recently, similar criteria were applied to CAD and promising results were obtained for the classification of breast lesions. The conventional 2D ultrasound of the breast is increasingly used in surgical clinical practice because it offers many benefits compared with other medical imaging techniques. Nevertheless, conventional 2D ultrasound images are not enough to transmit the entire US information of a solid breast lesion, while the 3D ultrasound can offer comprehensive information of all 2D lesion aspects and provide, in addition, simultaneously, the coronal plane. This additional information has been proved to be helpful for both clinical applications and CAD. The 2D CAD and 3D CAD are expected to be a useful CAD tool for classifying benign and malignant tumors in ultrasonograms, and can provide a second reading to help reduce misdiagnosis.

When mammograms are analyzed by computer, the pectoral muscle should be excluded from processing that is intended for the breast tissue. For this and other reasons, it is important to identify and segment out the pectoral muscle. In Chapter 18, a new, adaptive algorithm is proposed to automatically extract the pectoral muscle on digitized mammograms; it uses knowledge about the position and shape of the pectoral muscle on mediolateral oblique (MLO) views. The pectoral edge is first estimated by a straight line, which is validated for correctness of location and orientation. This estimate is then refined using iterative �cliff detection� to delineate the pectoral margin more accurately. Finally, an enclosed region, representing the pectoral muscle, is generated as a segmentation mask. The algorithm was found to be robust to the large variations in appearance of pectoral edges, to dense overlapping glandular tissue, and to artifacts like tape. The algorithm has been applied to the entire Mammographic Image Analysis Society (MIAS) database. The segmentation results were evaluated by two expert mammographic radiologists, who rated 83.9% of the curve segmentations to be adequate or better.

Image processing algorithms are the foundation for designing systems such as CADe to automatically process mammograms for tasks such as identifying potential abnormalities. A major limitation in the design of image processing algorithms lies in the difficulty in demonstrating that algorithms work to an acceptable measure of performance. Chapter 19 explores some of the issues, successes, and short comings related to existing performance evaluation paradigms for image processing algorithms in mammography, particularly in the context of image enhancement, segmentation, and registration. It reviews common algorithms in each category, the role that mammogram databases play in the performance evaluation of algorithms, and some of the challenges of designing image processing for mammograms.

Chapter 20 presents a simple, fast, and accurate method for automatically locating the nipple on digitized mammograms that have been segmented to reveal the skin-air interface. If the average gradient of the intensity is computed in the direction normal to the interface and directed inside the breast, a sudden and distinct change in this parameter close to the nipple is found. A nipple in profile is located between two successive maxima of this parameter; otherwise, it is near the global maximum. Specifically, the nipple is located midway between a successive maximum and minimum of the derivative of the average intensity gradient; these being local turning points for a nipple in profile, and global otherwise. The method has been tested on 24 images, including both oblique and cranio-caudal views, from two digital mammogram databases. For 23 of the images (96%), the rms error was less than 1 mm at image resolutions of 400 micrometers and 420 micrometers per pixel. Because of its simplicity, and because it is based on both the observed behavior of mammographic tissue intensities and on geometry, this method has the potential to become a generic method for locating the nipple on mammograms.

Automated detection of microcalcifications in mammograms is particularly challenging because microcalcifications are small and have low contrast. Typically, an automated system for detection of microcalcifications includes algorithms for segmentation, feature extraction, and classification that perform separation of candidate objects from the background, representation of candidate objects with informative metrics, and decision making, respectively. In Chapter 21, sequential process segmentation has a seminal role and dictates the performance of the entire system. Candidate objects must be delineated appropriately to allow effective feature extraction and classification. This chapter presents algorithms that specialize in segmentation of small, low- contrast objects, and illustrates their application to segmentation of microcalcifications.

Noninvasive assessment of microcirculatory characteristics of tissues using MRI is one of the emerging technologies as discussed in Chapter 22. Dynamic contrast-enhanced (DCE) breast magnetic resonance imaging (MRI), in which the breast is imaged before, during, and after the administration of a contrast agent, enables a truly 3D examination of breast tissues. This functional angiogenic imaging technique provides a noninvasive assessment of microcirculatory characteristics of tissues in addition to traditional anatomical structure information. The use of DCE breast MRI is increasing and it is recommended as an adjunctive breast screening modality to mammography. Traditional manual interpretation of DCE breast MRI is time- consuming and tedious and can lead to oversight errors due to the large size of 4D data sets (three spatial dimensions plus time). Manual interpretation is also subject to inter- and intra-observer variability. There is a great need for computer-aided detection and diagnosis systems capable of increasing the efficiency, accuracy, and consistency of breast MRI interpretation. In this chapter, the underlying principles of DCE breast MRI are reviewed the challenges of registration, segmentation, and classification in the design of CADe/CADx systems for breast MRI discussed, the progress made in this arena to date explored, and future directions for research suggested.

The pre- and postcontrast Gd-DTPA MR images of any body organ hold diagnostic utility in the area of medicine, particularly for breast lesion characterization. The Chapter 23 reviews state-of- the-art tools and techniques for lesion characterization, such as uptake curve estimation (functional segmentation), image subtraction, and velocity thresholding, differential characteristics of lesions such as maximum derivative of image sequence, steep slope and washout, fuzzy clustering, Markov random fields, and interative deformable models such as live- wire.

Research has shown that 10 to 30% of cases missed by mammography can be detected using breast MRI (BMRI). And while BMRI has increased sensitivity, it suffers from limited specificity. Chapter 24 BMRI, which is more difficult to interpret than mammography because it generates significantly more data, and there are fewer people qualified to use it for diagnosis because it is not the standard breast imaging modality. Mammographers appear to be the best qualified, and certainly the most likely to read BMRI, but they are accustomed to reading 4�6 images per patient as compared with the 200�400 images per patient that BMRI generates. Also, some mammographers may be uncomfortable interpreting MR images. A user-friendly computerized system for BMRI presentation and lesion classification may help mammographers overcome their objections and anxieties concerning BMRI. Such a system would reduce the time required to read BMRI, making the technique a more practical clinical tool. Research shows that the diagnostic accuracy of mammography is increased when two radiologists view the same mammogram, or when the same radiologist rereads a mammogram. Using such strategies decreases the number of missed cancers, which leads us to the fundamental reason of using CAD systems�that is, to replace the second pair of human eyes with a set of �electronic eyes.� The development of dependable, low-cost CAD systems for breast lesion detection and classification is of great practical interest. In this chapter, examples of CAD systems are shown that were developed to provide computer assistance for humans reading BMRI data and classifying BMRI lesions.

Before looking at a mammogram in detail, a radiologist scans it globally to ensure that it is adequate and of acceptable quality. Adequacy means that all of the relevant breast tissue has been captured on the mammogram. Image quality covers a multitude of attributes like exposure, sharpness, etc. Adequacy and image quality ensure that a subsequent decision on the presence or absence of an abnormality may reasonably be based on the information presented by the mammogram. With computer-aided diagnosis, it is equally important to discard inadequate, poor quality images before they are analyzed further, because such analysis would be futile. The automatic assessment of mammogram adequacy and quality is thus an important quality assurance step in computer-aided diagnosis of mammograms. Chapter 25 describes how aspects of image adequacy and quality may be assessed automatically by computer using guidelines established by radiologists themselves.

Visual inspection in medical care has always been a tedious and costly task that is prone to errors. Because of these errors, computerimaging systems have been designed to assist the human inspection process in the provision of better health care. Chapter 26 presents the analysis of the economic feasibility of any proposed system: costs associated with equipment operation, human inspection, false positives, and false negatives must be taken into account. Practical mathematical models based on probability theory can be used effectively to assist the decision-maker in the purchase of a computer-assisted imaging system by predicting its operational and financial performance implications. This chapter describes a mathematical model for the economic analysis of a computer-assisted imaging system for the detection of breast cancer.

Breast imaging is currently limited by inadequate sensitivity and specificity. About one in five breast cancers is missed by mammography screening, and when suspicious lesions are found and referred to biopsy, about 4 in 5 biopsies turn out to be benign and were thus arguably unnecessary. Chapter 27 addresses sensitivity, computer-aided detection has come a long way, with multiple commercial products now available and promising results from large clinical studies. On the other hand, specificity remains a major challenge, and to date there are still no commercially viable options due to medicolegal risks, lack of information explaining the computer diagnoses, and difficulties in clinical trial design. This chapter reviews the considerable research to date in computer-aided characterization in breast imaging, describes those current problems, and proposes some possible new approaches that may lead to eventual clinical translation of this important technology.

In Chapter 28, the book is concluded in contemplation of the future of computer aided detection. The future work that needs to be done to make the system more robust is discussed, as well as what it takes to develop a product, stages for FDA approval, and clinical trials. Finally, the chapter projects this research into the next decade.

Jasjit S. Suri, Ph.D.
Rangaraj, M. Rangayyan, Ph.D.


© SPIE. Terms of Use
Back to Top