Abstract: Automated MRI brain tumor segmentation provides a powerful tool for diagnosis. In this paper, a tumor segmentation method using multi-feature extraction and AdaBoost feature selection is presented. The method utilizes the information of MR images and the anatomical knowledge, and takes the advantage of the feature selection ability of AdaBoost. Experimental results on 20 slices MR images demonstrate the effectiveness of the feature selection, and achieve an accuracy of 96.82% on tumor segmentation.

Key words: Magnetic Resonance Image(MRI), tumor segmentation, texture features, feature selection, AdaBoost

摘要： 对脑部磁共振图像中肿瘤的自动分割，有助于了解疾病特征和制定手术方案，评价治疗效果。该文通过提取基于灰度统计、对称性、纹理等的特征，结合AdaBoost方法，利用计算机进行自动脑肿瘤分割。该方法综合利用了磁共振(MR)各加权图像的信息和大脑解剖结构的知识，以及AdaBoost算法的特征选择能力。在20帧带有肿瘤的MR图像上进行实验，得到了96.82%的分类准确率。

关键词: 磁共振图像, 肿瘤分割, 纹理特征, 特征选择, AdaBoost方法

CLC Number: 

• TP391.41