基于脑部MR图像GMM特征决策分类的肿瘤诊断

Diagnosis of Tumor in Brain MR Images Based on GMM Features and Decision Tree Classifier

磁共振(MR)图像提供了大量用于医疗检查的信息。精确鲁棒的脑部MR图像分割、特征提取和分类对于临床诊断肿瘤是非常重要的。提出一种新的基于脑部MR图像的肿瘤诊断方法。首先,通过多阈值分割形态学操作检测图像的畸形区域,然后,提取用于分类的高斯混合模型(GMM)特征,最后,利用决策树分类器对肿瘤图像类型进行分类。整个分类过程分为训练和测试2个阶段,训练阶段提取肿瘤图像和非肿瘤图像不同的特征,在测试阶段基于知识库进行肿瘤和非肿瘤分类。使用准确度、误报率和漏检率3个性能指标对算法进行评估,实验结果表明,分类准确度可达91.18%-94.11%,误报率和漏检率在2.94%-4.41%范围内,可以有助于更好的脑部肿瘤诊断。

Magnetic resonance imaging （MRI） offers a lot of information for medical examination. Accuracy and robustness of MRI segmentation, feature extraction and classification are very important for the diagnosis of tumor from the clinical point of view. A new tumor diagnosis method based on MR image of the brain is proposed. An abnormal area is detected using the multi-threshold segmentation with morphological operations of MRI firstly. Then, Gaussian mixture model （GMM） features are extracted for the purpose of classification. Finally, the decision tree classifier is used to classify the type of tumor image. The whole classification process is divided into stages of training and testing. In the training stage, various features are extracted from the tumor and non-tumor images. In the testing stage, based on the knowledge base, the classifier classifies the image into tumor and non-tumor. Accuracy and the false alarm and missed detection rate are used to evaluate the algorithm, experimental results show that the algorithm has excellent performance which can contribute to a better diagnosis of brain tumor.