科研之友
微信
新浪微博
Facebook
分享链接
摘要
In the diagnosis of adrenal tumor, the tumor's shape and clinical statistics in the tumor region are two main diagnostic bases, and the precise segmentation of tumors is important. The level set method has been proved effective in Computed Tomography image segmentation. However, it needs a manually delineated initial contour and highly depends on its accuracy. In this paper, a novel automatic method based on the sparse representation is proposed for the tumor segmentation. Firstly, image patches containing the tumor contour are used as the training set of the K-means singular value sparse decomposition algorithm to acquire an overcomplete dictionary sensitive to patches with edges. Secondly, the CT image is divided into overlapped patches and with the trained dictionary the sparse representation is used to identify patches containing the tumor contour. Then a region growing strategy is deployed to obtain an initial contour of the tumor. Lastly, this initial contour is applied to a localized region-based level set method (LRLSM) to obtain a precise tumor contour. The proposed method is compared with another normal method that applies manually delineated ellipsoidal initial contour to the LRLSM. The experimental results indicated that the proposed method obtained higher accuracy (90.31%) and lower mean absolute distance (0.997). Thus it is an accurate and reliable method and greatly improves the automaticity of the adrenal tumor segmentation.
- 出版日期2015-12
- 单位复旦大学
