Pattern Classification Methods for Analysis and Visualization of Brain Perfusion CT Maps - Computational Intelligence Paradigms in Advanced Pattern Classification - page 145

Chapter 8
Pattern Classification Methods for Analysis and
Visualization of Brain Perfusion CT Maps
Tomasz Hachaj
Pedagogical University of Krakow, Institute of Computer Science and
Computer Methods, 2 Podchorazych Ave, 30-084 Krakow, Poland
e-mail: [e-mail]
Abstract. This chapter describes basis of brain perfusion computed tomography
imaging (CTP) and computer based method for classification and visualization
perfusion abnormalities. The solution proposed by author - perfusion abnormality
detection measure and description (DMD) system - is consisted of the unified
algorithm for detection of asymmetry in CBF and CBV perfusion maps, the image
registration algorithm based on adaptation of a free form deformation model and
the description / diagnosis algorithm. The DMD system was validated on set of 37
triplets of medical images acquired from 30 different adult patients (man and
woman) with suspicious of ischemia / stroke. 77.0% of tested maps were rightly
classified and the visible lesions were detected and described identically to radiol-
ogist diagnosis. In this chapter the author presents also portable augmented reality
interface for visualization of medical data capable to render not only perfusion CT
data but also volumetric images in real time that can be run on off - the - shelf
computer .
1 Introduction
This chapter describes basis of brain perfusion computed tomography imaging
(CTP) and computer based method for classification and visualization perfusion
abnormalities. Author presents a set of advanced image processing and classifica-
tion methods that enables automatic interpretation of visualized symptoms.
Despite the fact that the solution is mainly dedicated to physician working in
neuroradiology departments it is also a good example of usage of advanced pattern
classification and computational intelligence heuristics in modern computer-aided
diagnostic (CAD) systems. Methods presented in this chapter (especially image
registration and tracking based augmented reality (AR) visualization) can be also
adapted to others image based pattern classification tasks not necessarily in the
field of medicine.