局部特征与多示例学习结合的超声图像分类方法

利用全局特征对超声图像进行描述具有一定的局限性,而且对图像进行手工标注的成本过高, 为解决上述问题,本文提出了一种利用局部特征描述超声图像,并结合多示例学习对超声图像进行分类的新方法. 粗略定位图像中的感兴趣区域 (Region of interest, ROI),并提取局部特征,将感兴趣区域看作由局部特征构成的示例包, 采用自组织映射(Self-organizing map, SOM)的方法对示例特征进行矢量量化,采用Bag of words方法将示例特征映射到示例包空间,进而采用传统的支持向量机对示例包进行分类.本文提出的方法在临床超声图像上进行了实验,实验结果表明,该方法具有良好的泛化能力和较高的准确性.

Combining Local Features and Multi-instance Learning for Ultrasound Image Classification

The method to describe ultrasound images using global features has some limitation. And it has a high cost to manually annotate the region of interest (ROI). To solve the above problems, local features are used to describe ultrasound images. A multi-instance learning method for ultrasound image classification is proposed. The ROI is roughly located and local features are extracted. The ROI is considered as a bag which is composed of local features. The self-organizing map (SOM) method is used for vector quantization and the bag of words method is used to map instance features to bag features. Then the classical support vector machine is used to classify the bags. The method proposed by this paper is tested by clinical ultrasound images. The results show that the method has a good generalization ability and has a high accuracy.