甲状腺结节超声图像特征提取及识别

Ultrasound image feature extraction and classification algorithm for thyroid nodules

采用多特征融合方法鉴别甲状腺结节超声图像的良恶性。首先用多尺度LBPV模型提取结节的局部纹理特征,然后与Tamura模型提取的全局纹理特征相结合,从全局和局部两方面对甲状腺结节的纹理特征进行了细致的描述。又提取了纵横比、圆形度、紧致度等形状特征,将上述特征进行融合并利用主成分分析法PCA对融合后的特征进行降维。实验结果表明,模型提取的特征用于分类识别时,较上述任一单一模型所提取特征用于分类时能获得更高的识别率。

Multi-feature fusion method is applied to distinguish benign and malignant thyroid nodules. First, multi-scale Local Binary Pattern Variance （LBPV） model is used to extract the local texture features of the nodules and then combined with the global texture features extracted with Tamura model to deseribethe texture characteristics of thyroid nodules globally and locally. Also the aspect ratio, circularity and compactness are extractedand fused, and then reduced the dimensionality with Primary Component Analysis （PCA）. Experimental results indicate that the proposed fusion model is more effective than any other single one.