基于磁共振影像特征集成融合的AD诊断

AD diagnosis based on integrated fusion of MR image features

为了得到更高更稳定的阿尔茨海默病(AD)诊断准确率,对脑磁共振影像纹理特征进行了集成融合,并用于AD分类诊断.首先,基于病理知识提取脑磁共振影像中左右脑相关解剖结构的体积、纹理特征;然后,采用链式智能体遗传算法与支持向量机相结合的封装式特征选择分类集成模型,对提取的特征集进行特征选择,从而实现融合;最后,利用融合后的特征进行分类诊断,并将融合后的分类结果与融合前以及采用p值法特征选择的分类结果进行对比.实验结果表明,相比融合前的特征以及采用p值法进行选择的特征,利用所提算法融合后的特征具有更高且更稳定的分类准确率、灵敏度和特异度.

In order to obtain higher and more stable diagnostic accuracy of Alzheimer＇s disease（ AD）,the texture features of magnetic resonance（ MR） images were integrated and fused for AD diagnosis. First,the volume and texture features of the left and right parts of multiple anatomical structures were extracted based on pathological knowledge. Secondly,by combining the chain-like agent genetic algorithm（ CAGA） and support vector machine（ SVM）,a feature selection classification ensemble model was designed to conduct deep feature selection and realize feature fusion. Finally,the fused features were used for classification and diagnosis of AD and the classification results are compared with those before fusion and those obtained by the p-value method. The experimental results showthat the features fused by this proposed algorithm have higher and more stable classification accuracy,sensitivity and specificity than the features before fusion and the features selected by the p-value method.