Medical imaging is an indispensable tool of patients’ healthcare in modern medicine. Machine leaning plays an essential role in the medical imaging field, including Computer-Aided Diagnosis (CAD), medical image analysis, organ/lesion segmentation, image fusion, image-guided therapy, image annotation, and image retrieval. It is generally difficult to derive analytic solutions or simple equations to represent objects such as lesions and anatomy in medical images because of large variations and complexity. For example, a lung nodule is generally modeled as a solid sphere, but there are nodules of various shapes and nodules with internal inhomogeneities, such as spiculated and ground-glass nodules. A polyp in the colon is modeled as a bulbous object, but there are also polyps that have a flat shape. Thus, tasks in medical imaging require “learning from examples” for accurate representation of data and incorporation of prior knowledge. Because of its essential needs, machine learning for/in medical imaging becomes one of the most promising, growing fields. As medical imaging has been advancing with the introduction of new imaging modalities and methodologies such as cone-beam/multi-slice Computed Tomography (CT), Positron-Emission Tomography (PET)-CT, tomosynthesis, diffusion-weighted Magnetic Resonance Imaging (MRI), electrical-impedance tomography, and diffuse optical tomography, new machine-learning algorithms and applications are demanded in the medical imaging field.
A large number of researchers participated in the development of a number of machine-learning methods in medical imaging, such as support vector machines, statistical machine-learning methods, manifold-space-based methods, artificial neural networks, and pixel-based machine learning. They applied or developed new ways of using these machine-learning methods for solving specific problems in medical imaging. However, there has been no book which unified these efforts. This book provides the first comprehensive overview of cutting-edge machine-learning research and technologies in medical decision-making based on medical images. This book covers major technical advancements and research findings in the field of CAD. Leading researchers in the field contributed chapters to this book in which they describe their state-of-the-art studies in machine learning in CAD and medical image analysis. This book demonstrates that such technologies and studies have reached the practical level, and that they are rapidly becoming available in clinical practice in hospitals.
The target audience of this book includes professors in engineering schools and medical schools, graduate and college students in engineering and applied-science departments, medical students, engineers in medical companies, researchers in industry, academia, and health science, medical doctors such as radiologists, cardiologists, and surgeons, as well as healthcare professionals such as radiology technologists and medical physicists. Professors may find this book useful for their classes. Engineers can refer to the book during their development of products. Researchers can use the book for their work and cite it in their publications. Medical doctors, medical physicists, and other health professionals can use the book to learn about state-of-the-art studies on machine learning in CAD.
This book consists of five sections organized by body parts: Section 1 deals with Breast Imaging, Section 2 with Thoracic Imaging, Section 3 with Abdominal Imaging, Section 4 with Brain Imaging, and Section 5 with Body Imaging. The book starts with breast imaging, because the first commercial product of Computer-Aided Detection (CADe) approved by the U.S. Food and Drug Administration was one for breast imaging. The first section contains five chapters provided by leading researchers on computer-aided diagnosis in breast imaging.
In the first chapter (Chapter 1) of the breast imaging section (Section 1), Drs. Gruszauskas, Drukker, and Giger describe a novel, important concept, the repeatability of computer-aided diagnosis (CADx). For CADx to be able to aid radiologists, a CADx system needs to demonstrate not only good, but also consistent performance. The authors have proposed the concept of repeatability to measure the consistency and robustness of the system performance. This concept can be one of the key factors in the successful development of a CADx system, and it is applicable to any CAD system.
The second chapter (Chapter 2) describes the computerized detection of architectural distortion, which is an early sign of breast cancer as seen on mammograms. Drs. Banik, Rangayyan, and Desautels have developed a CADe system for detection of architectural distortion in prior mammograms. This study is very important, because early detection of tumors increases the chance of patients’ survival. This chapter provides the recent research findings as to how much early detection of breast lesions is possible with the current state-of-the-art technology.
The third chapter (Chapter 3) reviews CADe and CADx of lesions in 3D breast imaging, provided by Drs. Muralidhar, Bovik, and Markey. 3D x-ray imaging is a novel form of breast imaging, including tomosynthesis, breast CT, and stereo mammography. 3D x-ray breast imaging provides a precise means of imaging breast lesions. This chapter provides reviews of the recent developments in 3D x-ray breast imaging and CAD systems for these modalities.
In the fourth chapter (Chapter 4), content-based image retrieval in CADx of breast lesions in mammography is described by Drs. El Naqa, Oh, and Yang. Relevance feedback used in content-based image retrieval is one of the keys to successful applications of image-retrieval systems to medical decision-making. This chapter presents recent advances in relevance feedback and its usefulness and potential in computer-aided medical decision-making.
The fifth chapter (Chapter 5) deals with fuzzy-set-theory-based methods in CADe and CADx of masses in mammography, provided by Drs. Sahba, Venetsanopoulos, and Schaefer. Because the inherent fuzziness in the nature of mammographic images makes fuzzy theory a useful technique for handling of these images, fuzzy-based methods are one of major technologies used in breast CADe and CADx.
The second section contains three parts on thoracic imaging, which is one of the major CAD areas. Because lung cancer is the leading cause of cancer deaths in the U.S. and other countries, detection and diagnosis of lung nodules are major topics in CAD research. Three leading groups in this area contributed chapters to this section.
In the first chapter (Chapter 6) in the thoracic imaging section (Section 2), Drs. Chen and Suzuki describe CADe of lung nodules in chest radiography. As chest radiography is the most frequently used diagnostic imaging modality for chest diseases, the impact of the research on lung nodule detection in chest radiography on healthcare is very large. The authors have developed a novel bone-separation technology based on pixel-based machine learning, called a massive-training artificial neural network, for improving the performance of a CADe system. Separation between bones and soft tissue improves the sensitivity for the detection of nodules overlapping ribs and clavicles, and it reduces false positives due to ribs and clavicles.
The second chapter (Chapter 7) in the second section is on automated segmentation of the lungs on CT images, provided by Drs. Sensakovic and Armato. Lung segmentation is the first indispensable step in CADe systems for the lungs. This chapter discusses common difficulties with lung segmentation, and it presents an advanced automated lung segmentation method which overcame these difficulties.
The development and actual use of a CADe system in a hospital, provided by Drs. Masutani, Nemoto, Numora, and Hayashi, is the subject of the third chapter (Chapter 8) in the second section. In particular, the study discussed in this chapter explores the issue of the training of a CADe system with new cases in a hospital. In this situation, the characteristics of the CAD system have to be adjusted to fit the dynamic changes in the patient population. The study provides one solution to this very important problem.
The third section contains four chapters on abdominal imaging for colon and prostate cancers. These areas are very important in healthcare, because colon cancer is the second leading cause of cancer deaths in the U.S. and other countries, and prostate cancer is the most common malignancy in American men. Leading researchers in the field contributed to this section.
The first chapter (Chapter 9) in the abdominal imaging section (Section 3) covers CADe of polyps in CT colonography, provided by Drs. Xu and Suzuki. CT colonography is a diagnostic imaging examination for screening of colorectal cancer with less invasiveness, less patient discomfort, and less examination time than occurs with conventional colonoscopy. CADe systems for polyps in CT colonography are demand because the sensitivity of inexperienced radiologists for CT colonography is low. The chapter describes feature selection and false-positive reduction with pixel-based machine learning. Feature selection is a major factor that determines the performance of the classifier in a CADe system, and the reduction in the number of false positives is crucial for a CADe system to be of practical use in hospitals.
The second chapter (Chapter 10) in the third section describes content-based image retrieval in CADe of polyps in CT colonography, provided by Drs. Yao and Summers. Content-based image retrieval is gaining attention as a useful means for improving the performance of a CADe system as well as that of radiologists. The authors developed a content-based image-retrieval method for CADe of polyps in CT colonography. This chapter reviews the recent developments of content-based image-retrieval methods, including their usefulness.
In the third chapter (Chapter 11) of the third section, a Bayesian method in CADe of polyps in CT colonography is presented by Drs. Ye and Slabaugh. They have developed a Bayesian method for improving the sensitivity and specificity of a CADe system for detection of polyps in CT colonography. The study described in this chapter demonstrates the effectiveness of the Bayesian method in a CADe system for polyp detection.
The fourth chapter (Chapter 12) in the third section is on CADe of prostate cancer on microscopic images of prostate tissue sections, provided by Drs. Peng, Jiang, and Yang, who developed an automated method for detecting prostate cancer from histologic images as an aid to pathologists. Because microscopic images are becoming digital, image analysis on and CADe with these digital histologic images are being used more and more in research. The study described in this chapter is one of the earliest attempts in this area.
The fourth section contains five chapters on brain imaging. Braining imaging is becoming more and more important in neurology and psychiatry, because it provides a less invasive, more quantitative means for detecting, diagnossis, and assesssment of brain diseases such as brain tumors, aneurysms, Alzheimer's disease, and autism. Modalities for brain imaging include T1-weighted MRI, T2-weighted MRI, contrast-enhanced MRI, diffusion-weighted MRI, and diffusion tensor MRI. Five groups leading in this area contributed chapters to this section.
The first chapter (Chapter 13) in the brain imaging section (Section 4) covers CADe systems for detection of brain diseases such as asymptomatic unruptured aneurysms, Alzheimer's disease, vascular dementia, and multiple sclerosis by MRI, provided by Drs. Arimura, Tokunaga, Yamashita, and Kuwazuru. Occlusions due to diseases and those of anatomic structures in 3D brain images can always occur because of the complex 3D structure of the brain. This may often prevent physicians from accurate detection, diagnosis, and assessment of brain diseases. CADe systems are expected to help physicians in these tasks during their reading of brain images.
The second chapter (Chapter 14) in the fourth section describes a computer-aided system for MRI in assessment of the therapeutic response of brain tumors to drug treatment, provided by Huo, Brown, and Okada. Computer-aided tumor response assessment is a promising area of the application of CAD technologies. With the computer aid, the response of tumors to drug treatment can be assessed more quantitatively and precisely, and thus, more precise, effective treatment planning is possible.
In the third chapter (Chapter 15) in the fourth section, CAD of autism by using 3D shape analysis in MRI is described by Drs. Elnakib, Casanova, Gimel’farb, and El-Baz. They have developed a computer-aided system for autism diagnosis based on analysis of the corpus callosum. The system provides a quantitative means for assessing the corpus callosum, and it will be useful in assisting physicians in the diagnosis of autism.
The topic of the fourth chapter (Chapter 16) in the fourth section is CADe of Alzheimer’s disease in brain MRI, provided by Drs. Zhang and Fan. This chapter reviews machine-learning techniques in CADe of Alzheimer’s disease as used with brain MRI from the point of view of neuroimage-analysis methods. CADe systems will be useful for early detection of Alzheimer’s disease.
The fifth chapter (Chapter 17) in the fourth section covers manifold learning techniques in medical image analysis, provided by Aljabar, Wolz, and Rueckert. Manifold learning is one of the most active areas in machine learning. It is expected that manifold learning will be able to represent medical data in a meaningful way. This chapter reviews manifold learning applications in image registration, segmentation, classification, and CADe.
The fifth section contains three chapters on CAD and image-guided radiation therapy in body imaging. Body imaging as described in this section includes torso CT, whole-body MRI, abdominal ultrasound, endoscopy, and kV and mV radiography. Leading researchers in the fields contributed to this section.
The first chapter (Chapter 18) in the body-imaging section (Section 5) describes manifold learning in CAD of lesions in various organs such as the brain, liver, and lungs in various modalities including whole-body MRI, brain MRI, CT, and ultrasound imaging, provided by Drs. Mateus, Wachinger, Atasoy, Schwarz, and Navab. This chapter presents manifold-learning applications to visualization, clustering, classification, registration, and human-motion modeling.
The second chapter (Chapter 19) in the fifth section deals with automated organ localization in torso CT, provided by Drs. Zhou and Fujita. Organ localization is a fundamental technique for identification of the locations of target organs. The authors have developed an automated method for this task based on ensemble learning. The automated-localization method provides fundamental information on target organs for image segmentation, lesion detection, content-based image retrieval, anatomic annotation, and CAD.
In the third chapter (Chapter 20) in the fifth section, machine learning in real-time tumor localization in radiation therapy is covered by Drs. Li and Jiang. The domain knowledge and prior information on tumor localization can be acquired with machine-learning techniques. These techniques are useful for improving the accuracy and robustness of tumor localization in radiation therapy of lung cancer. Machine learning is becoming more and more important in image-guided radiotherapy.
In summary, this book provides the first comprehensive overview of cutting-edge machine-learning research and technologies in CAD and medical image analysis. It covers major technical advancements and research findings in the field. Leading researchers contributed chapters describing their state-of-the-art studies. They describe recent developments and findings in their studies. In the studies, they applied and/or developed new ways to use machine-learning methods for solving specific problems in CAD and medical image analysis. Thus, readers can learn and gain knowledge from the book on machine learning in CAD and medical image analysis. Readers will know from the book that machine-learning technologies and studies have reached a practical level, and that they are becoming available in clinical practice in hospitals. Therefore, it is expected that researchers, professors, students, and other professionals will gain valuable knowledge from the book, use the book as a reference, and expand the current state-of-the art technologies. I hope that this book will inspire readers and help to advance the field of machine learning in CAD and medical image analysis.