ABSTRACT
One of the less explored machine learning methodologies in breast imaging CAD is instance-based learning, more commonly known as case-based reasoning (CBR) [3]. In contrast to learning methods that construct an explicit
Georgia D. Tourassi and Maciej A. Mazurowski
6.1 Introduction 95 6.2 e Case-Based CAD System Framework 96
6.2.1 e General Architecture 96 6.2.2 e Key Components 97
6.2.2.1 Case Base 97 6.2.2.2 Similarity Assessment 98 6.2.2.3 Search Engine 99 6.2.2.4 Decision Engine 99
6.3 Case-Based CAD Applications in Breast Imaging 100 6.3.1 Detection 100 6.3.2 Diagnosis 100 6.3.3 Other Applications 101
6.4 Hot Topics 101 6.4.1 Which Cases Should Be Included in the Case Base? 101 6.4.2 What Is the Most Eective Way to Assess Similarity? 103 6.4.3 What about Perceptual Similarity? 104 6.4.4 Optimizing the Case Base Search 105 6.4.5 Optimizing the Decision Algorithm 105
6.5 Concluding Remarks 106 Acknowledgments 106 References 106
model of the domain knowledge from the example cases, CBR systems (also known as case-based systems) simply store the examples as they become available. In that respect, case-based systems are not real learners, but procrastinating inference engines. When presented with a query case, a case-based system reviews the stored examples, assesses the query’s relatedness to them, and determines the most probable query label accordingly. Case-based CAD systems in medical imaging are related to another popular technology, content-based image retrieval (CBIR) [4]. Given a query image, case-based CAD systems operate initially as CBIR systems to identify similar example images according to the low-level visual content of the images. However, casebased CAD systems go one step further. ey subsequently apply an inference algorithm to provide a high-level semantic interpretation for the query using the retrieved examples.
