Medical Image Analysis and Informatics

Computer-Aided Diagnosis and Therapy

Paulo Mazzoncini de Azevedo-Marques, Arianna Mencattini, Marcello Salmeri, Rangaraj M. Rangayyan, Paulo Mazzoncini de Azevedo-Marques, Arianna Mencattini, Marcello Salmeri, Rangaraj M. Rangayyan

• Publisher: CRC Press
• Year: 2017
• ISBN:
  9781351230827
• Edition: 1
• Topic: Medicine
• Subtopic: Biotechnology in Medicine

With the development of rapidly increasing medical imaging modalities and their applications, the need for computers and computing in image generation, processing, visualization, archival, transmission, modeling, and analysis has grown substantially. Computers are being integrated into almost every medical imaging system. Medical Image Analysis and Informatics demonstrates how quantitative analysis becomes possible by the application of computational procedures to medical images. Furthermore, it shows how quantitative and objective analysis facilitated by medical image informatics, CBIR, and CAD could lead to improved diagnosis by physicians. Whereas CAD has become a part of the clinical workflow in the detection of breast cancer with mammograms, it is not yet established in other applications. CBIR is an alternative and complementary approach for image retrieval based on measures derived from images, which could also facilitate CAD. This book shows how digital image processing techniques can assist in quantitative analysis of medical images, how pattern recognition and classification techniques can facilitate CAD, and how CAD systems can assist in achieving efficient diagnosis, in designing optimal treatment protocols, in analyzing the effects of or response to treatment, and in clinical management of various conditions. The book affirms that medical imaging, medical image analysis, medical image informatics, CBIR, and CAD are proven as well as essential techniques for health care.

1

Segmentation and
Characterization of White
Matter Lesions in FLAIR
Magnetic Resonance
Imaging

1.1Introduction

1.2Background

MRI Fundamentals • Fluid Attenuation Inversion Recovery (FLAIR) MRI

1.3Challenges of Segmenting FLAIR MRI

Acquisition Noise • Intensity Inhomogeneity • Scanning Parameters • Intensity Non-Standardness • Partial Volume Averaging (PVA)

1.4Framework for Exploratory Noise Analysis on Modern MR Images

Testing for Stationarity • Testing for Common Distributions • Testing for Spatial Correlation

1.5Standardization and Brain Extraction

1.6PVA Quantification and WML Segmentation

The PVA Model • Edge-Based PVA Modeling • Fuzzy Edge Model • Global Edge Description • Estimating α • WML Segmentation

1.7Shape Analysis

Boundary-Based Techniques • Global Shape Metrics

1.8Results

Evaluation Metrics • Experimental Data • Exploratory Noise Analysis • Standardization • WML Segmentation Evaluation • Shape Characterization

Conclusion

References

Brittany Reiche Jesse Knight Alan R. Moody April Khademi

1.1Introduction

Acute ischemic stroke is described as the sudden interruption of blood flow to the brain that results in the deprivation of oxygen and nutrients to the cells; and stroke duration directly increases the risk of permanent brain damage. According to Statistics Canada, a government agency commissioned with the production of statistics to analyze all aspects of life in Canada, strokes were the third leading cause of death in Canada in 2011 (Statistics Canada, 2011), and they represent a 3.6 billion dollar a year burden on the economy in associated health costs and lost wages (Public Health Agency of Canada, 2011).

Physicians are now looking at Magnetic Resonance Images (MRI) to identify precursors to strokes. There is a strong relationship between white matter lesions (WML) and risk of stroke, as well as correlations with Alzheimer’s disease (Oppedal et al., 2015), multiple sclerosis (Grossman and McGowan, 1998), and vascular dementia (Hajnal et al., 1992). It has been noted that the prevalence of WML increases with age, and that the lesions are more common and extensive in those who already have cardiovascular risk factors or symptomatic cerebrovascular disease. WML are best seen in Fluid-Attenuated Inversion Recovery (FLAIR) MR images, manifesting as hyperintense objects distributed throughout the white matter, and this imaging modality has enhanced discrimination of ischemic pathology (Malloy et al., 2007). The total volume of these lesions are an important prognostic indicator for stroke risk (Altaf et al., 2006).

Traditionally, WML volume measurements are obtained by manual delineation; however, this is known to be laborious and subject to inter- and intra-observer variability. Automated image analysis techniques are a better alternative as they can segment WML accurately, efficiently, and consistently (Khademi et al., 2012). These methods are also ideal for large databases, as images can be processed quickly and without user intervention, in a way not feasible with manual processing. This is particularly important because technological advances have given way to the consolidation of large image repositories for multi-center studies. By analyzing this quantity of data, results will have more statistical significance and power (Suckling et al., 2014). However, due to the multi-center nature of this data, there is greater variability in image quality, contrast, and resolution. Methods developed for automatic segmentation must be able to account for these variations in order to be robust.

Many automatic segmentation methods have already been developed and generally fall into two categories: model-based or nonparametric. Model-based approaches tend to use intensity-based pixel classification with the Expectation–Maximization (EM) algorithm (Santago and Gage, 2003; Cuadra et al., 2002), where the model is constructed using a Gaussian Mixture Model (GMM). The results from these techniques are promising; however, they are based on the assumption that the underlying intensity distributions are Gaussian and also require estimates of distribution parameters. These assumptions lead to inaccurate segmentations in images from multi-coil MR scanners, as intensity distributions may be non-Gaussian and/or nonstationary (Khademi et al., 2009a). Also, the signal values of pathologies, like WML, do not follow a known distribution and cannot easily be handled by model-based approaches. Nonparametric techniques attempt to overcome the use of these assumptions by processing co-registered, multi-modality datasets (i.e., T1, T2, PD, FLAIR) (Anbeek et al., 2004; Lao et al., 2006; de Boer et al., 2007) to perform segmentation. These modalities are subsets of the MRI modality, where the resultant images have varying contrast qualities based on different parameter settings at image acquisition. This eliminates the requirement of assumed distributions, but increases the cost of image acquisition (multiple modalities per patient), computational complexity, and introduces registration error, reducing the appeal of this approach.

Current manual analysis of WML...

Table of contents

• Cover
• Half-Title
• Title
• Copyright
• Dedication
• Contents
• Foreword on CAD: Its Past, Present, and Future
• Preface
• Acknowledgements
• Editor bio
• Contributors
• 1 Segmentation and Characterization of White Matter Lesions in FLAIR Magnetic Resonance Imaging
• 2 Computer-Aided Diagnosis with Retinal Fundus Images
• 3 Computer-Aided Diagnosis of Retinopathy of Prematurity in Retinal Fundus Images
• 4 Automated OCT Segmentation for Images with DME
• 5 Computer-Aided Diagnosis with Dental Images
• 6 CAD Tool and Telemedicine for Burns
• 7 CAD of Cardiovascular Diseases
• 8 Realistic Lesion Insertion for Medical Data Augmentation
• 9 Diffuse Lung Diseases (Emphysema, Airway and Interstitial Lung Diseases)
• 10 Computerized Detection of Bilateral Asymmetry
• 11 Computer-Aided Diagnosis of Breast Cancer with Tomosynthesis Imaging
• 12 Computer-Aided Diagnosis of Spinal Abnormalities
• 13 CAD of GI Diseases with Capsule Endoscopy
• 14 Texture-Based Computer-Aided Classification of Focal Liver Diseases using Ultrasound Images
• 15 CAD of Dermatological Ulcers (Computational Aspects of CAD for Image Analysis of Foot and Leg Dermatological Lesions)
• 16In Vivo Bone Imaging with Micro-Computed Tomography
• 17 Augmented Statistical Shape Modeling for Orthopedic Surgery and Rehabilitation
• 18 Disease-Inspired Feature Design for Computer-Aided Diagnosis of Breast Cancer Digital Pathology Images
• 19 Medical Microwave Imaging and Analysis
• 20 Making Content-Based Medical Image Retrieval Systems Worth for Computer-Aided Diagnosis: From Theory to Application
• 21 Health Informatics for Research Applications of CAD
• Concluding Remarks
• Index