Chapter 6 - Handling of Feature Space Complexity for Texture Analysis in Medical Images

Abstract

In this chapter, we focus on the issue of feature space complexity when performing texture classification in medical images. In particular, to obtain accurate image classification, we would expect clear feature space separation between images of different classes. This means that ideally there should be low intraclass variation and interclass ambiguity in the feature space. However, such expectation is rarely met in real applications and the classification performance is thus affected. While the majority of existing studies focus on designing more representative and discriminative feature descriptors, we suggest that designing a classifier model that explicitly addresses the feature space complexity is an alternative direction in research. We provide a comprehensive review of such classification methods, including ensemble classification, subcategorization, and sparse representation. We also describe in detail a specific design on subcategory-based ensemble classification, with application in tissue pattern classification of Interstitial Lung Diseases (ILD) in High-Resolution Computed Tomography (HRCT) images. Some experimental results are presented as well.