Anke Meyer-Baese

Neuro-Fuzzy Classification

In a medical imaging system, uncertainties can be present at any point resulting from incomplete or imprecise input information, ambiguity in medical images, ill-defined or overlapping boundaries among the disease classes or regions, and indefiniteness in extracting features and relations among them. A patient can have a set of symptoms, which can be attributed to several diseases. The symptoms need not be strictly numerical. Thus, fuzzy variables can be both linguistic and/or set variables. This chapter reviews the basic concepts of fuzzy sets and the definitions needed for fuzzy clustering, and it presents several of the best-known fuzzy clustering algorithms and fuzzy learning vector quantization. Applications of neuro-fuzzy classification to medical image compression and exploratory data analysis are also shown.

Chapter 2 – Feature Selection and Extraction

Feature extraction methods encompass, besides the traditional transformed and nontransformed signal characteristics and texture, structural and graph descriptors. The feature selection methods described in this chapter are the exhaustive search, branch and bound algorithm, max–min feature selection, sequential forward and backward selection, and also Fisher’s linear discriminant. Advanced feature representation methods are becoming necessary when it comes to dealing with the local image content or with spatio-temporal characteristics or with the statistical image content. A review of the most important feature selection and extraction techniques for biomedical image processing is given.

Statistical and syntactic pattern recognition

This chapter gives an overview about the most important approaches in statistical and syntactic pattern recognition and their application to biomedical imaging. Parametric and nonparametric estimation methods and binary decision trees form the basis for most classification problems related to bioimaging while grammatical inference and graphical methods are the basic classification paradigms in syntactic pattern recognition. The chapter also reviews the diagnostic accuracy of classification measured by ROC-curves, and presents application examples based on statistical classification methods.

Chapter 3 – Subband Coding and Wavelet Transform

New transform techniques that specifically address the problems of image enhancement and compression, edge and feature extraction, and texture analysis received much attention in recent years especially in biomedical imaging. In this chapter, we review the basics of subband coding and present in great detail the different types of wavelet transforms.

Computer-Aided Diagnosis for Diagnostically Challenging Breast Lesions in DCE-MRI

Breast cancer is the most common cancer among women, but has an encouraging cure rate if diagnosed at an early stage. Thus, early detection of breast cancer continues to be the key for effective treatment. The success of CAD in conventional X-ray mammography motivated the research of automated diagnosis techniques in breast MRI to expedite diagnostic and screening activities. This chapter describes some important CAD systems for diagnostically challenging breast lesions in breast MRI. A CAD system for small lesion detection using integrated morphologic and dynamic characteristics and one for non-mass-like-enhancing lesions based on spatio-temporal features is described.

Spatio-Temporal Models in Functional and Perfusion Imaging

This chapter covers spatial approaches for three different types of temporal models: linear, nonlinear, and nonparametric models. Assuming a global spatial smoothness is typically not appropriate for medical images and locally adaptive smoothing allows to retain sharp features and borders in the images.

Chapter 12 – Analysis of Dynamic Susceptibility Contrast MRI Time-Series Based on Unsupervised Clustering Methods

The analysis of perfusion MRI data by unsupervised clustering methods provides the advantage that it does not imply speculative presumptive knowledge on contrast agent dilution models, but strictly focuses on the observed complete MRI signal time-series. In this chapter, the applicability of clustering techniques is demonstrated as tools for the analysis of dynamic susceptibility contrast MRI time-series and the performance of five different clustering methods is compared for this purpose.

Chapter 8 – Transformation and Signal-Separation Neural Networks

Neural networks are excellent candidates for feature extraction and selection, and also for signal separation. The underlying architectures are mostly employing unsupervised learning algorithms and are viewed as nonlinear dynamical systems. The material in this chapter is organized in three parts: the first part describes the neurodynamical aspects of neural networks, the second part deals with principal component analysis (PCA) and with related neural networks, and part 3 deals with independent component analysis (ICA) and neural architectures performing signal separation.

Chapter 4 – The Wavelet Transform in Medical Imaging

In this chapter, we present the theory of 2-D discrete wavelet transforms, and of biorthogonal wavelets, and we show several applications of the wavelet transform in medical imaging. The most remarkable applications are: (a) ability of the WT to make visible simple objects in a noisy background, which were previously considered to be invisible to a human viewer, (b) demonstrated superiority of the WT over existing techniques for unsharp mask enhancement and median filtering, and (c) enhancing the visibility of clinically important features.

Chapter 5 – Genetic Algorithms

Genetic algorithms (GA) like neural networks are biologically inspired and represent a new computational model having its roots in evolutionary sciences. In this chapter, we review the basics of GAs, briefly describe the schema theorem and the building block hypothesis, and describe feature selection based on GAs, as one of the most important applications of GAs.