Segmentation of PET/CT images by Flexible Mixture Models and Comparison with K-means and Gaussian Mixture Models

• Main Authors: Ye, Meng-Ciao, 葉孟樵
• Other Authors: Lu, Horng-Shing
• Format: Others
• Language: en_US
• Online Access: http://ndltd.ncl.edu.tw/handle/22461754161363238586

碩士 === 國立交通大學 === 統計學研究所 === 97 === Positron Emission Tomography (PET) helps doctors determine the abnormal regions. The specific brightened regions in PET images show the location of abnormal region. Hence the segmentation of the data form PET images is very important. There are three methods to classify the data from PET image to obtain the region of interest, K-means with KDE, Gaussian mixture model (GMM) with KDE and flexible mixture model (FMM) with KDE. The main difference between GMM and FMM is that, GMM uses several normal distributions to fit the original PET data, while FMM does not. FMM considers the property and structure of PET image data. It uses a right-skewed distribution to fit the background images. The mixture normal distribution is used to fit other regions. Finally, the result of FMM with KDE is better than the result of K-means with KDE and GMM with KDE.