Abstract: In order to realize image registration and fusion of thyroid tumor B-type ultrasound image and radionuclide image and get features areas which registration of the characteristic class required, it is necessary to segment thyroid, the thyroid tumor and surrounding tissue. Some spot noises make the image poor quality in the process of formate ultrasonic images and the image is characterized with low contrast and intensity inhomogeneity. An improved normalized cut segmentation method is proposed. Anisotropic diffusion is introduced into the normalized cut, avoiding the problem of over-and incomplete segmentation, denoising and enhancing edge for thyroid tumor ultrasonic image by adjusting the model parameters, optimizing edge figure and weight matrix. The experimental results confirm the feasibility.

Key words: image segmentation, B-type ultrasound image, thyroid tumor, normalized cut, anisotropic diffusion

摘要： 为了实现甲状腺肿瘤B超图像和核素图像的图像配准融合以及得到特征级图像配准所需要的特征区域，需分割出甲状腺、肿瘤及其周围组织。这类图像在形成过程中往往会产生斑点噪声使图像质量较差，且具有灰度对比度低和亮度分布不均匀等特点，提出一种基于各向异性扩散的归一化割分割法，将各向异性扩散模型引入到归一化割中，并通过调节模型参数来对甲状腺肿瘤B超图像进行去噪和边缘增强，优化了归一化割中轮廓线图和权值矩阵，在一定程度上避免了归一化割的过分割和欠分割，实验结果证明了该方法的可行性。

关键词: 图像分割, B超图像, 甲状腺肿瘤, 归一化割, 各向异性扩散