海马与胼胝体3D纹理分析在阿尔兹海默症诊断中的比较

Comparison between Hippocampus and Corpus Callosum 3D Texture Analysis in the Diagnosis of AD

目的利用脑MRI图像比较海马与胼胝体在阿尔兹海默病(AD)和轻度认知障碍(MCI)诊断中的差异。方法分别选取AD、MCI及正常对照(NC)患者各21例,分别提取其海马及胼胝体的三维纹理信息。通过方差分析选取特征参量,采取线性判别分析(LDA)和非线性判别分析(NDA)处理数据,利用反向传播(BP)神经网络模型对AD、MCI及NC进行分类识别,并比较海马与胼胝体纹理信息对分类结果的差异。结果 LDA和NDA两种方法分别对3类样本(AD、MCI、NC)进行分类判别的结果表明,海马的分类正确率均高于胼胝体;无论两组间(ADMCI、AD-NC、MCI-NC)还是三组间(AD、MCI与NC),无论训练集还是测试集,海马的分类识别正确率均高于胼胝体,其中海马的训练集分类正确率达到100%。结论利用三维纹理特征的神经网络模型可分类识别AD患者及MCI患者,并且海马的分类正确率高于胼胝体。

Objective To compare the differences between the hippocampus and corpus callosum in diagnosis of Alzheimer＇s disease（AD） and mild cognitive impairment（MCI） based on MR images. Methods Altogether 21 AD patients, 21 MCI patients and 21 normal controls（NC） were selected and three-dimensional texture of their hippocampus and corpus callosum was extracted. The characteristic parameters were selected through variance analysis. And the data was processed by using linear discriminant analysis（LDA） and nonlinear discriminant analysis（NDA）. Afterwards, a BP（Back Propagation） neural network model was built to classify and identify AD patients and MCI patients from NC. The different effects of hippocampus and corpus callosum texture in the classification results were compared. Results According to the classification and identification results of AD, MCI and NC by using LDA and NDA, hippocampus demonstrated higher classification accuracy than the corpus callosum; no matter among two groups（AD vs MCI, AD vs NC and MCI vs NC） and three groups（AD, MCI and NC）, and for the same training group and the testing group, hippocampus had higher classification accuracy than the corpus callosum. Hippocampus achieved 100% classification accuracy for the training group. Conclusion The neural network model using three-dimensional texture features can categorize patients with AD and MCI, and the classification accuracy of hippocampus is higher than that of corpus callosum.