基于支持向量机的核磁共振左心室图像自动检测与分割

基于支持向量机的核磁共振左心室图像自动检测与分割 SVM Based Auto Detection and Segmentation of the Left Ventricle MRI Image 屈颖歌 QU Ying-ge~1 1 2 3 4 first-author 周涛 ZHOU Tao~1 1 2 3 4 王平安 HENG Peng-an~2 1 2 3 4 夏德深 XIA De-shen~1 1 2 3 4 南京理工大学计算机系 南京理工大学计算机系 The Computer Department, Nanjing University of Science and Technology The Computer Department, Nanjing University of Science and Technology 香港中文大学计算机科学与工程系 香港中文大学计算机科学与工程系 Department of Computer Science and Engineering ,The Chinese University of Hong Kong, Hong Kong Shatin Department of Computer Science and Engineering ,The Chinese University of Hong Kong, Hong Kong Shatin 南京理工大学计算机系,南京,210094 南京理工大学计算机系,南京,210094 香港沙田 香港沙田 提出了用基于支持向量机的方法来实现核磁共振左心室图像的自动分割方法.首先用经过训练的支持向量机（SVM）在二维图像中进行识别和定位左心室目标区域并进一步找出边缘区域,采用一种改进的训练方法来提高SVM识别率,然后在足够准确的区域中利用梯度方法找出边缘点,并把他们连接起来,找出目标的边缘,达到分割的目的.实验表明,这种分割方法降低了SVM对背景图像的敏感度,提高了SVM识别率. A SVM based method is proposed aimed to perform the auto-segmentation of the left ventricle MRI image. The first and the most important step is to recognize and localize the target region that contains the left ventricle in the original two dimensional image with SVM. In this step an improved method is used to increase the recognize ratio. In the next step, a multi-localization thinking has been introduced to localize the edge areas, and a multi-level SVM is also introduced to improve the recognition ratio. Then within these exact regions, the gradient method is used to find the edge points and through connecting them to make out the margin of the target. The segmentation can also be preformed by other methods such as Level Set, and take the central region that can be auto formed through the recognized area in the first step as the initial contour. Experiment result proofs the advanced method, and the auto segmentation procedures do bring the following experiments lots of convenience. 支持向量机 磁共振成像（MRI）图像 自动检测 图像分割 support vector machine MRI image auto detection image segmentation TP274.4 香港特区政府研究资助局资助项目（CUHK/4180/01E） 2003-06-01 2021-04-01 6