基于弥散张量成像的卒中后抑郁风险预测模型

目的 探讨基于弥散张量成像（DTI）指标的改变所构建的卒中后抑郁（PSD）风险预测模型的临床价值。方法 回顾性分析2017年11月 - 2019年3月住院的101位脑卒中患者。上述101例患者入院行常规病史采集收集相关基线资料，待患者病情稳定后，行DTI检测感兴趣区（ROI）的FA值。脑卒中发病2月至半年间，每月对此101例卒中住院患者行汉密尔顿抑郁量表(HAMD)评分，将卒中患者根据评分结果分为卒中后抑郁(PSD)组和卒中后无抑郁(N - PSD)组。共计提取PSD组和N - PSD组基线资料和DTI影像学两大类特征参数22项，利用Lasso回归对这22个危险因素降维并筛选出独立危险因素后构建列线图（Nomogram）预测模型。对预测模型逐步进行内外校验，采用矫正曲线、受试者工作特征（ROC）曲线来评价模型的预测效能。结果 通过LASSO降维分别筛选出致PSD的2个基线资料危险因素（ADL评分、入院当天NIHSS评分），3个DTI影像学危险因素（左额叶FA、左颞叶FA、左前扣带回FA）。建立PSD的Nomogram预测模型并校验，矫正曲线的一致性测验发现，该PSD列线图的预测概率与实际概率具有良好的相关性。内部验证和外部验证的AUC分别为0.8535和0.8972。结论 基于DTI改变构建的PSD风险预测模型具有临床预测价值，有助于指导PSD的早期治疗并预防疾病发生发展。

Construction of a risk model associated with prediction of post-stroke depression based on diffusion tensor imaging

To develop a risk prediction model for post-stroke depression (PSD) based on diffusion tensor imaging(DTI). Methods Data of 101 stroke patients admitted to hospital (from November 2017 to March 2019) were retrospectivelyanalyzed. These 101 patients had been collected their routine medical history to collect relevant baseline data. After thepatients’ conditions being stable, the FA values of ROIs were measured by DTI. From the second month to the sixth month afterthe onset of stroke, these 101 patients were assessed by the Hamilton Depression Scale (HAMD) monthly. Based on the scores,patients were divided into a post-stroke depression (PSD) group and a post-stroke non-depression (N-PSD) group. Twenty-twocharacteristic parameters were extracted from the baseline data and the DTI imaging data. The least absolute shrinkage andselection operator (LASSO) regression was used to reduce the dimensions of these 22 risk factors and to screen out theindependent ones, then the Nomogram prediction model was established. The model’s predictive ability was validated bycalibration curve and the area under the curve (AUC) of receiver operating characteristic (ROC). Results Two demographiccharacteristics (the Activities of Daily Living score and the National Institutes of Health Stroke Scale score on the day ofhospitalization) and three DTI imaging risk factors (left frontal-lobe FA, left temporal-lobe FA, and left anterior cingulatecortex FA) were screened out by LASSO regression. The consistency test of the calibration curve found that there was a goodcorrelation between the predicted probability of the Nomogram for PSD and the actual probability. The AUC of internalvalidation and external validation were 0.8535 and 0.8972, respectively. Conclusion Based on DTI, the PSD risk model has aclinical predictive value, and it can help guide early treatment and prevent progression of PSD.