一种基于特征选择的医学图像检索方法

提出了一种基于特征选择的医学图像检索方法。考虑到医学图像的多类别特性，将分类与检索结合，采用AdaBoost方法对样本进行多次抽样，并将分类精度作为判据对特征进行选择，选取少量有利于分类的特征，同时将单特征弱分类器增强为强分类器。在检索阶段，本方法在选择后的特征子集以及类别子空间中进行检索。实验结果表明，与传统方法相比，本方法能达到较高的查准率，计算量也明显降低。

A Feature Selection Based Approach in Medical Image Retrieval

A novel method of medical image retrieval based on feature selection is proposed in this paper. Considering the multi-class property of medical images, it combines classification and retrieval parts, using AdaBoost method by re-weighting the select rate of training samples. Using classification accuracy as the criterion of feature selection, this method selects a few features from the feature pools which are propitious to retrieval; meanwhile, it boosts the weak classifiers into a strong one. In the retrieval step, this method retrieves medical images in the sub-feature set and sub-class space. Compared with general approaches, experiments show that, our method can achieve a high precision and a comparatively lower computation cost.