A Study on Object Contour Detection in the Medical Images

• Main Authors: Shys-Fan Yang-Mao, 楊茆世芳
• Other Authors: 朱延平
• Format: Others
• Language: en_US
• Published: 2007
• Online Access: http://ndltd.ncl.edu.tw/handle/08195328492093897015

博士 === 國立中興大學 === 資訊科學與工程學系 === 96 === Medical imaging designates the ensemble of techniques and processes used to create images of the human body or parts tissue for clinical purposes. There are several techniques to make those images, including the microscopy imaging, magnetic resonance imaging (MRI), computed tomography (CT), positron emission tomography (PET) or ultrasonography. The doctor can read these pictures to make diagnosis by analyze the locations or texture features of the objects in those medical images. However, if there are too many images that need to be analyzed, it is time consuming that all by manual. To speed up the time of analysis, it requires the computer science to automatic analysis those medical images. There are three main procedures in medical image analysis. The first procedure is image preprocessing, which is the image enhancement that includes image denoise and contrast enhancement. The second procedure is the region of interest (ROI) segmentation, also called object segmentation. The third procedure is the further object analysis, which is the medical knowledge based image analysis. The most important procedures are the first and second ones, the result of image enhancement will affect the accuracy of object segmentation; and the object segmentation required several tuning for different images. In this thesis, we focus on retinal fundus and cervical smear images. The main goal is locating the optical disc contours from fundus images and the nucleus or cytoplast contours from Pap-smear images. The accuracy of objects segmentation on those images is very sensitive to noise and poor image quality. To solve this question, we propose several automatic image segmentation techniques in this thesis. The experimental results show that all the techniques proposed here have performed impressively. Besides cervical smear images or retinal fundus images, these techniques can also be utilized in detecting object contours on other images.