多谱MR脑图象的组织分类

TISSUE CLASSIFICATION OF MULTI SPECTRAL MR IMAGES OF HUMAN BRAIN

传统的基于灰 度的 Ｍ Ｒ Ｉ 脑图象分 割很 难达 到临 床应 用的 需要， 因为 受 Ｒ Ｆ 线圈、 Ｍ Ｒ 设备 的操 作环境等的影响 ，在同一 幅 图象 中或 各 扫描 图象 间都 存 在灰 度不 均 匀性。 本 文采 用 聚类 分 析法 ，由 单谱图象的灰 度和纹理参数构 成特征向量，进 行初始分 类。在 此基础 上采用 适配 分割算 法，通 过估计有偏场， 对多谱图象的灰 度进行校正，从 而实现对脑组织 的正确分类。

Conventional intensity based segmentation of brain MR images can hardly meet the needs of the clinical application. Intra scan and inter scan intensity inhomogeneities due to RF coils and the operating conditions of the MR equipment are a common source of difficulty. In this paper, cluster analysis was used for the initial classification in the feature space consisting of intensity of a MR image and its textures. Then a new method called adaptive segmentation was performed. Using the expectation maximization (EM) algorithm, we iteratively alternated evaluations of a posterior tissue class probabilities and the bias field, and finally obtained correct segmentation of multispectral MR images.