This book is concerned with a machine learning approach to analyze and understand biomedical images. The development of medical imaging technologies in the last three decades has been truly revolutionary, while in parallel, machine learning has experienced a vertiginous advance in recent years. Several challenging problems in the biomedical domain and the set of powerful machine learning techniques have resulted in a new domain on its own, where the power and beauty of these techniques can be fully exploited to obtain proper solutions to these challenges. The main goal of the book is to highlight the great research potential of this interdisciplinary area, providing insights on new potential applications of machine learning techniques to the solution of important problems in biomedical image applications.
The prime intended audience of the book corresponds to researchers in both biomedical imaging and machine learning. Biomedical imaging researchers, as well as practitioners will find insights on how to adapt/apply machine learning techniques to tackle challenging image analysis problems. Also the book will provide useful material for machine learning researchers looking for interesting application problems. The book also provides material for supporting advanced undergrad and graduate courses on biomedical image analysis and/or machine learning.
The book presents a selection of 14 high-quality chapters, written by 43 authors, from eight different countries. The book is organized in 5 parts: Introduction, Feature Extraction, Machine Learning Based Segmentation, Biomedical Image Understanding and Interpretation, and Complex Motion Analysis, which can be considered as natural domains of applications for machine learning techniques.
The introduction part includes one chapter, "From Biomedical Image Analysis to Biomedical Image Understanding Using Machine Learning", written by the editors. The chapter presents an overview of the main topics covered by the book, emphasizing the fundamental concepts and techniques. The last part of the chapter focuses on the main problem in image analysis, image understanding, i.e., the problem of relating the low-level visual content of an image with its high-level semantic meaning. The chapters presents some promising research directions that provide insights on how to use machine learning to tackle with this problem.
Part II focuses on the feature extraction process, which is fundamental for any image analysis task. In the context of biomedical image analysis, feature extraction is particularly important since it facilitates the inclusion of problem specific knowledge in the process.
Chapter two, "Computer-Aided Detection and Diagnosis of Breast Cancer using Machine Learning and Texture and Shape Features", focuses on the problem of breast cancer diagnosis supported by computerized analysis of digital mammograms. The chapter discusses different techniques, giving especial attention to methods that use texture and shape features to characterize tissues.
Chapter three, "Machine Learning for Automated Polyp Detection in Computed Tomography Colonography", proposes two different features for codifying the shape characteristics of polyps, and non-polyps, in computed tomography colonography. The features are orientation independent and their calculation is not computationally demanding. The features are tested using different state-of-the-art machine learning algorithms, showing a good performance on polyp detection.
Part III is devoted to the problem of image segmentation using machine learning techniques. Image segmentation is one of the main problems in image analysis. In biomedical image analysis, segmentation has several applications such as localization of pathologies, organ extraction for morphometry analysis, and cell quantification in histology slides.
Chapter four, "Variational Approach Based Image Pre-Processing Techniques for Virtual Colonoscopy", addresses the problem of colon segmentation for computed tomographic colonography using a variational approach. This approach uses a statistical model for regions based on Gaussian functions with adaptive parameters, which are learned using maximum likelihood estimation. Finally, pixels are classified as tissue or non-tissue using a Bayesian classifier.
In Chapter five, "Machine Learning for Brain Image Segmentation", the authors cast image segmentation as a supervised learning problem in a Bayesian framework. The Chapter presents a new algorithm, AdaSVM, a method that combines AdaBoost, as a feature selection method, with a support vector machine classifier. The algorithm shows a competitive performance when compared to other state-of-the-art approaches for supervised brain image segmentation.
In Chapter six, "A Genetic Algorithm-based Level-set Curve Evolution for Prostate Segmentation on Pelvic CT and MRI Images", the authors propose a genetic algorithm for optimizing the parameters of a segmenting contour implicitly defined by a level-set. The genetic algorithms attempts to minimize an energy function associated to the level-set function. The algorithm is applied to the problem of prostate segmentation in Pelvic CT and MRI images.
Chapter seven, "Genetic Adaptation of Level Sets Parameters for Medical Imaging Segmentation", proposes an analogous method to the previous one. The main difference is that in this method the genetic algorithm is not used to directly adapt the parameters of the segmenting curve. Instead, the genetic algorithm is used to estimate the parameters of an algorithm that attempts to fit a Gaussian curve to the organ's slice histogram in order to model the level-set propagation speed. The method is tested with a liver segmentation task on computer tomography medical images.
Part IV is dedicated to the problem of understanding the image contents by structuring the biomedical knowledge with very different strategies. Automated extraction of biomedical knowledge is a challenging but necessary task in the current technological world, in which large amounts of information are available but not utilized.
In chapter eight, "Automatic Analysis of Microscopic Images in Hematological Cytology Applications", the authors explore a great variety of methods to detect, classify and measure objects in hematological cytology: the most relevant image processing and machine learning techniques used to develop a fully automated blood smear analysis system. Likewise, recent advances of main automated analysis steps are presented.
Chapter nine "Biomedical Microscopic Image Processing by Graphs" overviews graph-based regularization methods. These methods have been extended to address semi-supervised clustering and segmentation of any discrete domain that can be represented by a graph of arbitrary structure. These graph-based approaches are combined to attack various problems in cytological and histological image filtering, segmentation and classification.
In chapter ten, "Assessment of Kidney Function Using Dynamic Contrast Enhanced MRI Techniques" the kidney is segmented using level sets and then classified under three different metrics: Euclidean distance, Mahalanobis distance and Least Square support vector machine. Classification accuracy, diagnostic sensitivity, and diagnostic specificity result to be 84%, 75%, and 96%, respectively.
Chapter eleven, "Ensemble of Neural Networks for Automated Cell Phenotype Image Classification", is focused on the study of machine learning techniques for cell phenotype image classification and demonstrates the advantages of using a multi-classifier system instead of a stand-alone method to solve this difficult classification problem
Chapter twelve, "Content-Based Access to Medical Image Collections", describes state-of-the art techniques for accessing large collections of medical images, retrieving similar images to the examined one or visualizing the structure of the whole collection. Both strategies take advantage of image contents, allowing users to find or identify images that are related by their visual composition. In addition, these strategies are based on machine learning methods to handle complex image patterns, semantic medical concepts, image collection visualizations and summarizations.
Part V is devoted to the problem of motion analysis, which adds a time, dynamic dimension to image analysis and understanding. In this context, motion analysis is understood in two different and complementary senses: first, a user interacting with an image using an image visualization interface, second, structures changing through time in a sequence of images.
In chapter thirteen, "Predicting Complex Patterns of Human Movements using Bayesian Online Learning in Medical Imaging Applications", the authors present a Bayesian framework which is able to follow different complex user movements. The Bayesian strategy is implemented through a particle filter, resulting in real time tracking of these complex patterns. Two different imaged patterns illustrate the potential of the procedure: a precise tracking a pathologist in a virtual microscopy context and a temporal follow up of gait patterns.
Chapter fourteen, "Left Ventricle Segmentation and Motion Analysis in Multi-Slice Computerized Tomography" is concerned with the problem of cardiac motion estimation. A short overview of machine learning techniques applied to several imaging modalities is presented. This method is based on the application of Support Vector Machines (SVM), region growing and a nonrigid bidimensional correspondence algorithm used for tracking the anatomical landmarks extracted from the segmented left ventricle (LV). Some experimental results are presented and at the end of the chapter a short summary is presented.
Putting together a diverse set of contributions to constitute a coherent whole was a challenging task. But also it was an enriching and rewarding experience. We are really grateful with the contributors and reviewers, not only because of their outstanding work, but because of all the new and interesting things we learned from them. We are sure that readers will share with us this feeling.