目的 通过机器学习的方法分析脑拓扑属性数据,并探索精神分裂症患者大脑网络拓扑属性的改变.方法 2022年1月至2023年8月采集60例精神分裂症患者和56例健康对照者的功能磁共振影像数据并进行预处理,构建脑功能网络,提取全局和节点拓扑属性.将所有受试者划分为训练组与测试组.基于支持向量机对训练组数据进行拟合并通过交叉验证检验预测性能;通过递归特征消除算法优化模型,提取对预测性能做出最大贡献的指标.基于最优预测性能的训练模型计算测试组分类性能.采用SPSS 20.0进行统计分析,采用独立样本t检验和卡方检验比较两组数据差异.结果 支持向量机基于所有指标预测精神分裂症患者的测试组准确率为75.00％,在剔除冗余特征后结合递归特征消除算法-支持向量机模型测试组预测准确率提升至90.00％.模型中左侧额部颞上回、右侧背部无颗粒岛叶、双侧背部颗粒岛叶、双侧尾部扣带回以及左外侧前额丘脑的节点全局效率(nodal global efficiency,Ne)是对分类有最大贡献的特征,即相较于对照组,精神分裂症患者在以上脑区Ne值存在异常.结论 精神分裂症患者多个脑区Ne值存在异常,提示信息整合传递功能的异常可能与患者脑网络动态平衡失衡有关.
Characteristics of brain network topological properties in schizophrenic patients based on machine learning
Objective To analyze brain topological property data through machine learning methods and explore changes in brain network topological properties in patients with schizophrenia.Methods From January 2022 to August 2023,functional magnetic resonance imaging data of 60 patients with schizophrenia and 56 healthy controls were collected,and the data were preprocessed to construct brain functional networks and extract global and nodal topological properties.All subjects were divided into a training group and a tes-ting group.The data of training group were fitted based on support vector machine,and the predictive per-formance was evaluated through cross-validation.The model was optimized by recursive feature elimination al-gorithm,then the indicators that contributed the most to predictive performance were extrated.The classifica-tion performance of the testing group was calculated based on the trained model with optimal predictive per-formance.SPSS 20.0 software was used for data analysis,the independent t-test and X2 test were used for comparing the differences between the two groups.Results The support vector machine achieved an accura-cy of 75.00％in predicting the test group of schizophrenia patients based on all indicators.After removing redundant features and combining with the recursive feature elimination algorithm,the accuracy of the SVM model in predicting the test group increased to 90.00％.The nodal global efficiency(Ne)of the left superior temporal gyrus,right dorsal agranular insula,bilateral dorsal granular insula,bilateral caudal cingulate gyrus,and left lateral orbitofrontal cortex in the model contributed the most to classification.Compared to the control group,patients with schizophrenia had abnormal Ne values in these brain regions.Conclusion There are multiple brain regions with abnormal Ne values in patients with schizophrenia,indicating that the abnormali-ties in information integration and transmission functions may be related to the imbalance in the dynamic e-quilibrium of the patients'brain networks.
SchizophreniaFunctional magnetic resonance imagingGraph theoryMachine learningSupport vector machine
NETL
NSTL
万方数据
