海马MRI定量技术联合临床信息对认知衰弱风险的预测价值

MRI Quantitative Techniques in the Hippocampus Combined with Clinical Information in Predicting the Risk of Cognitive Decline

目的联合海马MRI定量指标和临床信息构建可逆性认知衰弱(RCF)风险因素预测模型。资料与方法回顾性收集遵义医科大学第三附属医院2020年1月—2022年12月RCF 41例和非RCF老年人40例,受试者均行头颅MRI定量磁化率成像(QSM)和三维动脉自旋标记扫描,通过标准化方法提取双侧海马区脑铁含量和脑血流量,同时收集临床信息构建基线表。先使用最小绝对收缩和选择算子回归和单因素Logistic回归筛选变量,再使用多因素Logistic回归分析筛选出独立预测变量进行模型构建并绘制诺模图,最后通过加强Bootstrap法500次重复抽样进行模型内部验证,使用区分度、校准度及决策曲线评价模型。结果RCF与非RCF组脑血流量、QSM值、社交活动、睡眠质量差异有统计学意义(χ^(2)=5.13、4.27、9.13、15.53,P均<0.05)。多因素Logistic回归分析筛选出睡眠质量、社交活动、QSM及脑血流量4个变量用于建模。模型预测RCF发生风险的区分度良好,曲线下面积为0.927(95%CI 0.856~0.978);校准曲线显示模型预测RCF发生风险与实际情况高度吻合(χ^(2)=52.20,P=4.14);决策曲线显示模型具有临床适用性。结论MRI定量数据联合临床信息建立的多尺度RCF临床预测模型具有较好的区分度、校准度及临床适用度,可为RCF风险因素筛查提供一定帮助。

Purpose To construct a risk factor prediction model for reversible cognitive frailty(RCF)by combining MRI quantitative indexes and clinical information of hippocampus.Materials and Methods Forty-one patients with RCF and 40 elderly people without RCF served as a control group were retrospectively included in this study.All subjects underwent quantitative susceptibility mapping(QSM)and three-dimensional arterial spin labeling scans to extract the cerebral iron content and blood flow indexes in bilateral hippocampus by standardized methods.Meanwhile,clinical information was collected to construct baseline tables.First,least absolute shrinkage and selection operator regression and single Logistic regression were used to screen variables,then multivariate Logistic regression analysis was used.The selected independent predictors were used to construct the model and draw Nomo chart.Finally,the internal validation of the model was carried out by strengthening Bootstrap method for 500 times of repeated sampling.The differentiation,calibration and decision curve were used to evaluate model.Results There were significant differences in cerebral blood flow value,QSM value,social activity and sleep quality between RCF and non-RCF groups(χ^(2)=5.13,4.27,9.13,15.53,all P<0.05).The independent risk factors screened by multivariate Logistic regression analysis were four variables:sleep quality,social activities,QSM and cerebral blood flow.The model had a good differentiation in predicting the risk of RCF,with area under curve was 0.927(95%CI 0.856-0.978).The calibration curve showed that the model predicting the risk of RCF occurrence was highly consistent with the actual situation(χ^(2)=52.20,P=4.14).The decision curve showed that the model had clinical applicability.Conclusion The multi-scale clinical prediction model of RCF based on MRI quantitative data combined with clinical information has good differentiation,calibration and clinical applicability,which can provide certain help for the screening of risk factors for RCF.