Parameter Optimization For Segmenting Structures In CT images

Abstract

Medical imaging plays an important role in the diagnosis, therapy and treatment of various organs, tumors and other abnormalities. It benefits the patient through more rapid and precise disease management with lesser side effects. Segmentation in medical imaging provides an important role for calculating the geometric shape and size of tumors and abnormal growth of any tissue. Automatic segmentation in medical imaging is the challenging job for the researchers. It automatically calculates the exact values of place, position and area of the tumor or structural part of the image, which is needed for surgery and other treatments. There are many techniques available for auto-segmentation of images like Active contours, Fuzzy based classifiers, Gradient Vector Field theory, Tensor based segmentation, Level set theory etc. But many of them are suffering from problems like optimization, initialization and insufficient results in noisy images. In this thesis we tried to optimize the Level set based segmentation process for different images on the basis of the texture analysis. In traditional methods, one has to select each parameter one by one to get best result and thus optimization is manually for every new image. However all the images have differed features like energy, entropy and contrast etc and these features are vary image to image (even within one image for its different parts). We tried to correlate the segmentation according to the texture features of an image, to make it automatic (no initialization of parameters) and more efficient