摘要
目的采用扩散张量成像(DTI)的自动纤维定量(AFQ)分析方法,使用支持向量机(SVM)模型方法对轻度认知障碍(MCI)的不同亚型进行分类预测研究以帮助对该病的鉴别诊断。方法搜集18例遗忘型轻度认知障碍(aMCI)患者、21例非遗忘型轻度认知障碍(naMCI)患者和25名健康对照组(NC)的临床及影像学资料,用AFQ软件包对所有被试的DTI数据进行分析,追踪全脑20条白质纤维束,每条纤维束分成100等份,对每个等份的各向异性分数(FA)值及平均扩散系数(MD)值定量分析,提取出三组被试的脑白质纤维特征,并将其作为分类特征,利用SVM对MCI进行分类预测,记录并比较其分类效力。结果 AFQ成功追踪获得的纤维束包括双侧皮质脊髓束、双侧丘脑放射束、双侧扣带束海马、双侧扣带束扣带回、胼胝体束膝部和压部、双侧下额枕束、双侧弓形束、双侧钩状束、双侧上纵束和双侧下纵束。与NC组相比,aMCI患者在右侧皮质脊髓束中间节段(节点48~59)、胼胝体束压部(节点1~8、31~50、53~64、74、82~87)FA值降低,左侧上纵束偏前部(节点11~44)、左侧下纵束的中间节段(节点33~48、56~63)及钩束的偏下部(节点66~78、97~100)MD值增高。与naMCI患者相比,aMCI患者左侧皮质脊髓束偏上部(节点87~100)FA值降低。所有纤维束所有节点的FA值作为分类特征在aMCI和其他两组分类中的具有综合更好的敏感性、特异性和总体准确率,且曲线下面积(AUC)值最大,为0.913。结论本研究基于DTI的AFQ分析构建的SVM模型获得较为精确的分类预测结果,对MCI的分类预测具有较高的应用价值。
Objective Using automatic fiber quantitative(AFQ) analysis of magnetic resonance(MR) diffusion tensor imaging(DTI),support vector machine(SVM) model was used to classify and predict different subtypes of mild cognitive impairment(MCI) in order to aid differential diagnosis of the disease. Methods Clinical and imaging data of 18 patients with amnestic mild cognitive impairment(aMCI),21 patients with non-amnestic mild cognitive impairment(naMCI) and 25 normal control(NC) were collected. The DTI data of all the subjects were analyzed by using AFQ software package.Twenty white matter fiber bundles were tracked in the whole brain, and each fiber bundle was divided into 100 equal sections. The FA and MD values of each equal section were quantitatively analyzed and the white matter fiber characteristics of three groups of subjects were extracted and used as classification characteristics. Support vector machine(SVM) was used to classify and predict MCI,and its classification effectiveness was recorded and compared. Results The fiber tracts successfully tracked by AFQ included bilateral corticospinal tracts, bilateral thalamic radiation, bilateral cingulum cingulate, bilateral cingulumhippocampus, callosumforceps, bilateral inferior frontal occipital fasciculus, bilateral superior longitudinal fasciculus, bilateral inferior longitudinal fasciculus, bilateral arcuate fasciculus and bilateral uncinate fasciculus.Compared with the healthy control group, FA values were decreased in the middle partof the right corticospinal tract and the splenium of corpus callosum, while MD values were increased in the anterior part of the left superior longitudinal fasciculus and the middle part of the left inferior longitudinal fasciculus and the lower part of the uncinate fasciculus. Compared with naMCI patients, the FA value of the upper part of the left corticospinal tract in aMCI patients was lower. As a classification feature, FA all showed better comprehensive sensitivity specificity and overall accuracy in aMCI and the other two groups, with the highest AUC value of 0.913. Conclusion In this study, the SVM model based on the AFQ analysis of DTI obtains relatively accurate classification prediction results andhas high application value for MCI classification prediction.
作者
戴珂
陆加明
李卫萍
张鑫
青钊
张冰
DAI Ke;LU Jiaming;LI Weiping(Department of Radiology,The Affiliated Drum Tower Hospital,Medical School of Nanjing University,Nanjing,Jiangsu Province 210008,P.R.China)
出处
《临床放射学杂志》
北大核心
2022年第1期23-29,共7页
Journal of Clinical Radiology
基金
国家自然科学基金项目(编号:81720108022)
江苏省社会发展科技项目(编号:BE2017707)
江苏省卫计委“科教强卫”工程医学重点人才项目(编号:ZDRCA2016064)
江苏省人社厅“六大人才高峰项目”高层次人才项目(编号:WSN-138)。
关键词
轻度认知障碍
扩散张量成像
自动纤维定量
支持向量机
分类
Mild cognitive impairment
Diffusion tensor imaging
Automatic fiber quantification
SVM
Classification
