EXPERIMENTS IN MEDICAL IMAGE SEGMENTATIONS
MetadataShow full item record
Non-invasive Radiology Imaging (e.g. CT, MRI, and PET) have been utilized tremendously in medical study for disease diagnosis, prognostication, and monitoring therapeutic response. And segmenting medical image for regions of interest is an essential step in computer assisted clinical interventions. Tumor detection in biomedical imaging is a time-consuming process for medical professionals and with nonneglectable human variation in recent decades, researchers have developed algorithmic techniques for image processing using a wide variety of mathematical methods, such as statistical modeling, variational techniques, and machine learning. Graph theory is the framework for the study explained in this thesis. We focus on both theoretical and practical aspects, with an emphasis on the experimental, practical parts. On the theoretical part, we propose a graph cut based semi-automatic method for liver segmentation of 2D CT scans into three labels denoting healthy, vessel, or tumor tissue. In chapter one, some definitions and algorithms for networks are introduced. Medical data and the idea of convolution are also introduced. In chapter two, we propose a new model to segment using the image sequences from dynamic CT scan. Also a particular image can be picked from the images sequence and then vectorized using a sequence of convolutions or neighborhoods. We also introduce a method of minimization that we call the α method. In chapter three, a moving mean method is developed and tested, using a more traditional train-test partition of the data. In chapter four, we explore the potential of deep neural networks. We decided to begin with U-Net, which consists of a contracting path to capture context and a symmetric expanding path that enables precise localization and we developed for medical data.