摘要： 计算机断层成像(CT)对疾病的确诊意义重大，在医学图像的自动检测中应用较多的模型为图割模型，但传统图割算法严重依赖于对复杂区域进行大量建立的模型，运算复杂且不利推广。为此，在传统图割理论基础上引入核函数，提出一种基于核图割模型的肝脏CT图像肿瘤分割算法。通过核函数将原始数据映射到高维空间，并在高维图像数据空间用图割理论对CT图像的肝区与肿瘤区域进行分割，以提取疑似肿瘤区域，解决传统图割模型中需要依赖人机交互和对复杂区域建模困难等问题。由Mercer定理得出，核空间的点积运算不需要显式指定图像各区域的具体模型，进行核推广后克服了传统模型通用性不强的弱点。利用临床CT图像数据对该算法进行分割实验，结果表明，基于核推广后的图割算法能够有效对肿瘤和肝区进行分离，可应用于临床实际中作为肿瘤辅助诊断手段。

关键词: 图割, 核图割, 肿瘤分割, 肝脏分割, 医学图像分割

Abstract: Computed Tomography(CT) images are significant in disease diagnosis, whereas graph cut model has been widely used in the automatic detection of complicated disease. Due to the fact that the complex area of medical images is very hard to model in conventional graph cut literature, this paper adopts the kernel trick in such a way that the segmentation of tumor is computed in the high dimensional kernel space rather than in the traditional spatial space directly. The processing of complex modeling and human-computer interaction is hereby avoided thought kernel trick. Moreover, Mercer’s theory proves that the computation of kernel method is implied and the model of different area is explicitly needless, which implies that the kernel graph cuts is universal to different applications. The proposed approach is validated on a real CT image data from clinical case, and the tumor is successfully extracted from the liver images. Results show that the proposed approach can be further ameliorated and applied to clinic as an auxiliary diagnosis assistant in the further.

Key words: graph cut, kernel graph cut, tumor segmentation, liver segmentation, medical image segmentation

中图分类号: 

• TP391.41