目的 基于多变量模式分析(multivariate pattern analysis,MVPA)对飞行学员和健康的普通人的大脑功能连接进行有效识别.材料与方法 采集了40名已经取得执照的飞行专业在校学生与39名地面专业在校学生的功能磁共振数据.通过网络功能连接分析得到功能连接矩阵作为特征,分别通过最小绝对收缩选择算子(least absolute shrinkage and selection operator,LASSO)算法与独立样本t检验方法对特征降维.使用不同核函数的支持向量机(support vector machine,SVM)进行训练和预测,使用留一交叉验证法进行模型性能评估,最终根据训练后SVM模型中的权重定位对应脑区之间的功能连接.结果 使用LASSO特征筛选的线性(linear)核SVM模型准确率为81.82%,敏感度82.05%,特异度81.58%,曲线下面积(area under the curve,AUC)为0.88.核函数对模型准确率的影响不大.模型中右侧中央旁小叶、双侧中央后回、双侧顶下缘角回、右侧梭状回、左侧眶部额中回、左侧顶上回、右侧眶部额下回有较高的权重,模型中的权重集中在感觉运动网络(somatomotor network,SMN)与默认模式网络(default mode network,DMN),分别占用所有权重的25.62%和25.27%.结论 结合LASSO算法进行特征筛选的SVM可以对飞行学员大脑进行有效识别,并且有更好的可解释性和更小的过拟合.模型权重信息反映了飞行学员主要在运动能力和感知能力有别于普通人.
Research on the recognition of brain functional connections in flight students based on multivariate pattern analysis
Objective:Based on multivariate pattern analysis(MVPA),effectively identify the brain functional connections between flight cadets and healthy individuals.Materials and Methods:Functional magnetic resonance data were collected from 40 licensed flight major students and 39 ground major students.The functional connectivity matrix was obtained through network functional connectivity analysis as a feature,and the feature dimensionality was reduced using the least absolute shrinkage and selection operator(LASSO)algorithm and independent sample t-test method,respectively.Support vector machines(SVM)with different kernel functions were used for training and prediction,and the performance of the model was evaluated using the left one cross validation method.Finally,the functional connections between corresponding brain regions were located based on the weight information in the trained SVM model.Results:The linear kernel SVM model using LASSO feature screening had an accuracy of 81.82%,sensitivity of 82.05%,specificity of 81.58%,and area under the curve(AUC)of 0.88.The kernel function had little effect on the accuracy of the model.In the model,the right paracentral lobule,bilateral posterior central gyrus,bilateral inferior parietal angular gyrus,right fusiform gyrus,left orbital frontal gyrus,left superior parietal gyrus,and right orbital inferior frontal gyrus had higher weights.The weights in the model were concentrated in the somatomotor network(SMN)and default mode network(DMN),accounting for 25.62%and 25.27%of all weights,respectively.Conclusions:SVM combined with LASSO algorithm for feature filtering can effectively recognize the brain of flight students,and has better interpretability and smaller overfitting.The weight information of the model reflects that flight students are mainly different from ordinary people in terms of motor and perceptual abilities.
万方数据
维普
