【摘要】 对大脑组织的影像分割,在人脑疾病的诊断和治疗以及人脑的功能研究中有着越来越重要的作用。本文应用最近邻算法(kNN)和模糊最近原型(FNP)算法对Multi-SpectralMR图像进行分割。综合考虑MR图像的特征和解剖学的知识,我们发展了一套邻域规则,用来弥补kNN算法和FNP算法的各自的不足,加快了计算的速度。通常的分类分析只在状态空间,而没有考虑到图像中像素之间的相邻信息,而正是这些由生物学特征决定的相邻信息,可以作为状态空间分类之后,用邻域规则对通常分类中不确定的像素进行进一步分类的依据。通过这样的这步过程,有效地降低了分类的错误,得到的分割结果明显好于文献中单独使用kNN和FNP算法得到的结果。更多还原

【Abstract】 Segmentation for brain tissue becomes more and more important in clinical diagnosis and treatment for diseases as well as for basic research such as for virtual human subjects. In this paper, pattern recognition methods were used to perform the segmentation for Multi-Spectral MR images. The k Nearest Neighbor (kNN) and the Fuzzy Nearest Prototype (FNP) algorithmswere adopted.Combination of the consideration on the feature of MR image and the anatomic knowledge, a set of neighborhood rules were established. Generally, kNN and FNP algorithms classify the samples in their state space, ignoring the information between pixels in the image. The neighborhood rules were added to determine the uncertain results after state space classification. The result from the segmentation was obviously better then that from previous kNN or FNP algorithm.更多还原