Comparative Studies of Contouring Algorithms for Cardiac Image Segmentation
Syed Farooq Ali · OhioLink ETD Center (Ohio Library and Information Network) · 2011
In the past few decades, cardiac image analysis especially segmentation has been a subject of several studies.In its basic form, cardiac image segmentation involves identifying heart or any of its anatomically or physiologically relevant features from 2D or 3D images.Methods for cardiac image segmentation can be divided into two major categories, namely, low and medium level cardiac segmentation techniques and high level cardiac segmentation techniques also called model based pattern recognition methods.The low and medium level cardiac image segmentation techniques can be further divided into three categories: (a) histogramming and thresholding based cardiac image segmentation techniques, (b) edge based cardiac image segmentation techniques and (c) mathematical morphological approaches for cardiac image segmentation.These approaches suffer from several drawbacks.For example, majority of these techniques are based on thresholding or histogramming and therefore disregard all local features in the image.Prior distribution is not utilized in these approaches.Moreover, these approaches do not include hand-drawn techniques of cardiac segmentation and hence, the concept of fitting mathematical equations to justify the approach does not exist.Model based pattern recognition methods include learning-based techniques and methods involving energy minimization.These methods can overcome the shortcomings iii encountered in low and medium level segmentation techniques.These approaches do not require thresholding and priors can be estimated from the measured images.The aim of this thesis is to not only review the previous work related to cardiac image segmentation but also implement and compare results from several state-of-the-art segmentation techniques, namely, cardiac contour initialization using morphological operators, cardiac segmentation using snakes (active contour) and its extensions, and STACS (stochastic active contour scheme).These approaches are described in more detail in Chapter 3.In conclusion, cardiac image segmentation remains an active area of research.Further developments, especially to improve robustness, are required to broaden the applicability of these techniques for medical imaging data.Further exploration of model based pattern recognition methods may provide more consistent and accurate results.