甲状腺结节超声图像多特征融合及识别

为更加精确地判别甲状腺结节的良恶性, 提出基于改进的 CLBP(Completed Local Binary Pattern)模型和 GLCM(Gray Level Co-occurrence Matrix)模型相结合的纹理特征提取算法。 首先在传统的 CLBP 模型中引入局部 方差信息, 使 CLBP 算子对局部纹理特征的描述更加精细; 然后与 GLCM 模型描述的全局纹理特征相融合; 最 后结合纵横比、 圆形度、 紧致度等形状特征并将其输入 SVM(Support Vector Machine)分类器。 为进一步提高识 别率, 同时给出基于粒子群算法与网格搜索算法相结合的 SVM 参数优化算法, 以提高识别率。 对比实验结果 表明, 该算法提取的特征用于分类识别时具有更高的识别率, 且提出的参数寻优算法相对于传统寻优算法效率 更高。

Multiple Feature Fusion and Recognition of Thyroid Nodule Ultrasound Image

In order to distinguish the benign or malignant thyroid nodules more accurately, we propose the texture feature extraction algorithm based on improved CLBP(Completed Local Binary Pattern)model and GLCM(Gray Level Co-occurrence Matrix)model.Firstly,the local variance information is introduced into the traditional CLBP model to make the CLBP operator describe the local texture features more precisely.Then,it is combined with the global texture features described by the GLCM model.Finally,the shape features such as aspect ratio, roundness and compactness are combined and input to the SVM(Support Vector Machine) classifier.In order to further improve the recognition rate, the SVM parameter optimization algorithm based on particle swarm optimization combining with grid searching is proposed.The experimental results show that the feature extracted by the algorithm presented in this paper has higher recognition rate for classification and recognition,and the proposed parameter optimization algorithm is more efficient than the traditional ones.