Department of Neurology, Tongji Hospital, School of Medicine, Tongji University, Shanghai 200092,China.
IFlytek Co., Ltd.,China.
Curr Alzheimer Res. 2021;18(14):1127-1139. doi: 10.2174/1567205018666211210150808.
This study aimed to build the supervised learning model to predict the state of cognitive impairment, Alzheimer's Disease (AD) and cognitive domains including memory, language, action, and visuospatial based on Digital Clock Drawing Test (dCDT) precisely.
207 normal controls, 242 Mild Cognitive Impairment (MCI) patients, 87 dementia patients, including 53 AD patients, were selected from Shanghai Tongji Hospital. The electromagnetic tablets were used to collect the trajectory points of dCDT. By combining dynamic process and static results, different types of features were extracted, and the prediction models were built based on the feature selection approaches and machine learning methods.
The optimal AUC of cognitive impairment's screening, AD's screening and differentiation are 0.782, 0.919 and 0.818, respectively. In addition, the cognitive state of the domains with the best prediction result based on the features of dCDT is action with the optimal AUC 0.794, while the other three cognitive domains got the prediction results between 0.744-0.755.
By extracting dCDT features, cognitive impairment and AD patients can be identified early. Through dCDT feature extraction, a prediction model of single cognitive domain damage can be established.
本研究旨在构建基于数字时钟绘制测试（dCDT）的监督学习模型，以精确预测认知障碍、阿尔茨海默病（AD）的状态和包括记忆、语言、动作和视空间在内的认知领域。
从上海同济大学附属医院选取 207 名正常对照者、242 名轻度认知障碍（MCI）患者和 87 名痴呆患者，包括 53 名 AD 患者。使用电磁板采集 dCDT 的轨迹点。通过结合动态过程和静态结果，提取不同类型的特征，并基于特征选择方法和机器学习方法构建预测模型。
认知障碍筛查、AD 筛查和鉴别诊断的最佳 AUC 分别为 0.782、0.919 和 0.818。此外，基于 dCDT 特征的预测结果最佳的认知域为动作，最佳 AUC 为 0.794，而其他三个认知域的预测结果在 0.744-0.755 之间。
通过提取 dCDT 特征，可以早期识别认知障碍和 AD 患者。通过 dCDT 特征提取，可以建立单个认知域损伤的预测模型。
