基于图割与粗糙集的MRI脑部肿瘤图像检索方法

背景：基于内容的医学图像检索是一门涉及多领域的学科，由于各种医学图像的成像原理不同，产生的图像在颜色、纹理和形状等视觉特征方面存在差别，使得此方法的实现还存在许多需要解决的问题。 目的：针对基于内容的医学图像检索中存在特征提取困难、检索时间长的问题，提出一种基于图割与粗糙集结合的相似图像检索方法。 方法：为克服图割仅适用于较少象素的图像和倾向于小割集的缺陷，首先对图像进行聚类，然后构建图像的Gomory-Hu割树，按割值大小依次去掉值较小的边，提取出图像的特征子图并构建特征库。为实现快速检索，借助粗糙集对特征库中的特征进行约简，有效减少参与相似性比较的特征数量。并将此方法应用到MRI脑部肿瘤图像的检索。 结果与结论：实验结果表明该方法能快速有效地检索出MRI脑部图像库中的肿瘤图像，检索的平均查准率为78.4%，平均查全率为62.9%。

Brain tumor image retrieval method based on graph cuts and rough sets

BACKGROUND: Content-based medical image retrieval involves multiple domains. Due to different imaging principles of various medical images, there are differences in color, texture, and shape, which should be resolved. OBJECTIVE: As in content-based medical image retrieval system, feature extraction from image is very difficult and the retrieval is very time-consuming, a similar image retrieval method based on graph-cuts and rough sets is proposed. METHODS: In order to overcome the defects that graph-cuts is only suitable for small image and easily leads to a small cut-sets, a clustering was applied to image, and the Gomory-Hu cuts tree of image was established. An image feature library was built by removing the edges of Gomory-Hu cuts tree for the value of cut. Reduction of features in library was obtained based on rough sets and the number of features in similar compare decrease. This method was applied to retrieve brain tumor image in MRI brain image database. RESULTS AND CONCLUSION: Results show that this method can effectively retrieve brain tumor images in the library. The average retrieval precision rate and the average recall rates were 78.4% and 62.9%, respectively.