基于眼动跟踪技术诊断阿尔茨海默病的Logistics回归模型的初步建立

A logistic regression model for diagnosing Alzheimer’s disease based on eye movement tracking technology

目的 初步构建基于眼动跟踪技术的阿尔茨海默病(Alzheimer’s disease,AD)诊断模型。方法 收集重庆医科大学附属第一医院老年科记忆门诊于2021年1-11月期间诊断为AD的患者113例,以及同期的认知正常体检者54例,应用眼动追踪技术对全部受试者进行眼动检查,测定静息态眼跳次数等共计5项指标。随机抽取70%病例(117例)分入训练集,剩余30%病例(50例)分入测试集。分析训练集病例注视眼跳次数等共计5项指标用于预测AD的价值,综合以上指标,再利用SPSS软件建立Logistic回归模型,使用ROC曲线评估模型的诊断效能,并在测试集中比较并验证模型对AD的诊断价值。分别将两组受试者间有显著性统计学意义的变量分别和自由延迟线索回忆(free and cued selective reminding test,FCSRT)、画钟试验(clock drawing task,CDT)、数字广度测试(digit span test,DST)和连线试验(trail making test,TMT)、听觉词语学习检测(auditory verbal learning test,AVLT)、波士顿命名测验(Boston naming test,BNT)的值进行相关性分析以探讨其可能的机制。结果 2组患者年龄、性别、受教育程度、注视眼跳次数、追随任务眼跳次数间的差异均无统计学意义。正向眼跳任务眼跳延迟时间、反向眼跳任务眼跳延迟时间、反向眼跳任务错误率间的差异均有统计学意义(P<0.001)。使用多因素Logistic回归构建包含了注视眼跳次数、反向眼跳任务延迟时间、反向眼跳任务错误率的预测模型,该模型预测AD的AUC在训练集和测试集中分别为0.913和0.964。试验组中反向眼跳延迟时间与FCSRT、TMT、AVLT测试分数呈负相关,反向错误率与FCSRT、CDT、AVLT、BNT测试分数呈负相关。结论 基于眼动跟踪技术检测指标构建的诊断模型对于AD的预测效果较好。

Objective To establish a logistic regression model based on eye tracking technology to diagnose Alzheimer’s disease(AD).Methods A total of 113 AD patients diagnosed in the clinic of memory of the Department of Geriatrics of the First Affiliated Hospital of Chongqing Medical University from January 2021 to November 2021 were recruited in this study,and another 54 individuals with normal cognition who taking physical examination during same period served as control group.All subjects were tested by eye tracker,and 5 indicators including saccades in the fixation task(resting state) were measured.Then,70%(117) subjects were randomly included in the training set;and the rest 30%(50) subjects in the test set.After the above 5 indicators were used to predict AD in the training set,a logistic regression model was established by using SPSS software,and receiver operating characteristic(ROC) curve was drawn to evaluate the diagnostic efficacy of the model,which was then testified in the test set.The diagnostic factors with statistically significant results between the 2 subject groups were correlated respectively with the scores of free and cued selective reminding test(FCSRT),clock drawing task(CDT),digit span test(DST) and trail making test(TMT),auditory verbal learning test(AVLT),and Boston naming test(BNT) so as to study the possible mechanism.Results There were no statistical differences between the 2 groups regarding age,gender,education level,saccades in the fixation task and saccades in the pursuit task.But significant differences were observed between the 2 groups in saccade latency in the pro-saccade task,saccade latency in the anti-saccade task and error rate of anti-saccade task(all P<0.001).A prediction model based on multivariate logistic regression was established and consisted of saccades in the fixation task,saccade latency in anti-saccade task and error rate of anti-saccade task in prediction of AD with an AUC value of 0.913 and 0.964 respectively in the training and test sets.In the experimental group,the anti-saccade latency was negatively correlated with the scores of FCSRT,TMT and AVLT,and the anti-saccade error rate was negatively correlated with the scores of FCSRT,CDT,AVLT and BNT.Conclusion Our established diagnostic model based on eye tracking data is of good prediction for the diagnosis of AD in clinical practice.