基于对称性理论的医学图像多阶段分类算法

摘要

A large number of medical CT images collected by PACS are widely used in clinical diagnosis.ROI and the features of ROI extracted from CT images can be utilized to classify these medical images so as to assist doctors to improve the efficiency and precision of the diagnosis. Brain imaging shows that it is approximately symmetrical about the brain stem.Based on this medical knowledge guidance,a medical image multi-stage classification (MSC)based on the theory of symmetry is presented in this paper.First of all,weak symmetry and strong symmetry is defined to describe the symmetry from the different granularities.Then,the weak symmetry decision algorithm was given to finish the first-stage classification for medical image in the coarse granularity.Further,the strong symmetry decision algorithm based on the point symmetry is proposed,combining with the weak symmetry decision algorithm,to complete the second-stage classification for the abnormal images classified from the first-stage in the fine granularity in order to locate lesion area.Finally,the features extracted from the lesions are used for the third-stage classification to help the doctor’s diagnosis.Experimental results show that multi-stage classifi-cation method based on the theory of symmetry can increase the accuracy of the classification and reduce the time of the doctor’s diagnosis.%利用影像归档和通信系统收集的大量医学 CT 图像被广泛应用于临床诊断，从中提取的 ROI 区域和 ROI区域的特征可以用来对医学 CT 图像进行分类，从而辅助医生提高诊断精度。医学图像的成像结果显示一张医学图像关于中垂线两侧近似对称。基于这一脑部医学领域知识的指导，文中提出了基于对称性理论的医学图像多阶段分类（Multi-Stage Classification，MSC）方法。首先，文中提出了弱对称性和强对称性的定义，从不同粒度对医学图像的对称性进行了描述；然后，给出了基于灰度直方图相交性的弱对称性判定算法，对医学图像在较粗粒度上进行了第1阶段的分类；接着，提出了基于点对称的强对称性判定算法，结合弱对称性判定算法，对第1阶段分类结果为异常的图像进行了更细粒度的第2阶段分类，定位了病变区域的位置；最后，利用对病变区域所提取的特征，对病变区域进行了第3阶段的分类，以达到辅助医生诊断的效果。实验结果表明，基于对称性理论的医学图像多阶段分类方法提高了医学图像分类的准确度，同时减少了医生诊断决策的时间。