Application of different transformation methods to whole heart region segmentation

This study presents the results of whole heart region segmentation and enhancement by using several transformation methods in cardiac magnetic resonance imaging (MRI). A Gaussian filter-based method, Fast Fourier Transformation (FFT), Laplacian of Gaussian (LoG), Discrete Cosine Transform (DCT), and Hadamard transformation are tested in order to analyze and overcome some of the problems that are faced during the heart segmentation. The first part of the study deals with the segmentation while the second part applies different transformation methods for analysis. Cardiovascular diseases are the leading cause of death in the world over the last decade. To reduce the mortality, early diagnosis is very important. Automatic segmentation and enhancement of cardiac structures can help physicians in various states of treatment of heart diseases. Manual tracing of heart region from cardiac MRI images is very difficult. Some of the methods proposed by previous studies include thresholding, probability theory, and deformable models. Thresholding based methods are not practical due to the existence of smooth boundaries in the cardiac images. Probability based methods require relatively large image database since it needs to adapt itself with the pathology. Deformable models are widely used for the segmentation problems; however they tend to fail in case of artifacts, diffused boundaries, and they do not use priori information which can be helpful for getting better segmentation results. Our work combines the deformable model with a traditional segmentation method. At first, the deformable model is used to locate the boundaries to avoid the complex process for locating the boundaries manually. Then, the edge information detected by Canny filter is used to modify diffused boundaries. Different transformation methods are applied to the segmented region to enhance and analyze the details.