Automatical Cardiovascular Calcification Detection from Chest CT Images

• Main Authors: Wen-Hsiang Chen, 陳文祥
• Other Authors: Hsi-Jian Lee
• Format: Others
• Language: en_US
• Online Access: http://ndltd.ncl.edu.tw/handle/37507990004811941741

碩士 === 慈濟大學 === 醫學資訊研究所 === 95 === Heart diseases are the third of the top ten causes of death in 2006 in Taiwan, according to the Department of Health, and it is also one of the main causes of death in developed countries. The common heart diseases include myocardial infarction, ventricular hypertrophy, and cardiovascular calcification. A preliminary examination of cardiovascular calcification uses computed tomography, which is non-invasive, safe, and simple. Although physicians have to examine patients’ computed tomography images, each patient has around 200 computed tomography images in an examination. The diagnostic efficiency deteriorates and subtle lesions may be ignored because physicians have to examine different diseases simultaneously in many computed tomography images. In this study, we will develop an algorithm to detect cardiovascular calcifications automatically. It is a challenging task using computed tomography to check cardiovascular calcifications. Arteries are difficult to locate in chest computed tomography images using image processing or pattern recognition methods, because arteries’ grayscales are very similar and the distance between arteries is short. Recently, many researchers proposed methods using single or multiple images to detect calcifications. They did not use differences between different images. In this thesis, we will detect calcifications according to information in consecutive images, called sequence information. The system consists of three parts: organs detection and tracking, cardiovascular region extraction, and calcification detection. In the first part, the body types of patients and the setting parameters of medical devices are different. Each patient’s image series are also different. According to human anatomy, the trachea and lungs in the chest computed tomographic images have very apparent appearances that are black regions. We can extract quickly cardiovascular regions by organs detection and tracking. The second part is for cardiovascular region extraction. Since the heart and cardiovascular vessels are located between the left and right lungs, we can extract cardiovascular regions efficiently. In the third part, calcification detection, we will use local thresholding, top-hat transform, and sequence information to detect calcifications. Because different grayscales show variant calcification degrees, we cannot use a fixed threshold to detect calcifications. However grayscales distributions of normal arteries’ are similar. We use a filter to analyze the color distribution. If the distributions have one or more sharp peaks and white regions, then we mark them as calcification regions. In the experiments, we applied our algorithm to 10 patients’ image sequences. The average number of images per patient is 229 and the average time for detecting cardiovascular calcifications is 159 seconds. The recall rate for calcification detection is 92.6%. The experiment results show that our algorithm can detect cardiovascular calcifications quickly and accurately.