Segmantation in abdominal medical images

Abdominal medical image segmentation is the process of labeling each voxel in a medical image dataset to indicate its tissue type or anatomical structure. Identifying organs from abdominal Computed Tomography (CT) datasets is one of the essential steps for surgical planning. However, due to gray level similarity of adjacent organs, injection of contrast material and partial volume effects; robust segmentation of abdominal organs is a very difficult task. In this thesis, segmentation of liver parenchyma, which is essential for live donor liver transplantation surgery, is studied. Three different automatic and semi-automatic hybrid liver segmentation algorithms are proposed and their performances are compared. All proposed algorithms have three stages; namely, preprocessing, classification/thresholding and post-processing. In the first algorithm, k-means clustering is used as a classifier. In the second algorithm, Multi Layer Perceptron, which is a more complex classifier, is used. In the third algorithm dynamical thresholding is used to identify liver parenchyma. The first algorithm is able to deal with contrast variations. In addition, it is worth to say that this algorithm is much faster supporting applicability for clinical usage. The second and third algorithms are able to deal with not only contrast variations but also atypical liver shapes. These capabilities are provided by the patient oriented structure which learns the characteristics of a patient dataset for each slice in parallel to the segmentation process and adapts its parameters according to these characteristics. Keywords : Liver Segmentation, Computed Tomography, Multi Layer Perceptron Network, k-means, Dynamical Thresholding