An intelligent screener for mild cognitive impairment via integrated eye-tracking and the digital clock drawing test

Background Mild cognitive impairment (MCI) is a risk factor for dementia, and early screening is crucial for patient prognosis. Objective To construct an intelligent family screening model for MCI based on eye tracking (ET) and digital clock drawing tests (dCDT), to provide a simple and accurate screening tool for MCI. Methods This study included 618 cognitively normal participants and 179 patients with MCI, among whom demographic information and metrics from ET and dCDT were collected. One-way analysis of variance was applied to screen all variables (p < 0.05). Different feature sets constructed based on logistic regression and five machine learning methods (random forests, multilayer perceptron, support vector machines, extreme gradient boosting trees, and convolutional neural networks) were used to construct 36 MCI screening tools. Finally, the diagnostic efficacy of the models was evaluated based on the area under the receiver operating characteristic curve (AUROC), accuracy, sensitivity, and specificity. Results Multimodal features, namely demographics, dCDT, and ET, showed superior performance compared to models based on unimodal behavioral data with or without demographics. Among all algorithms, the random forest model based on all significant features performed the best, with an AUROC of 0.947. Conclusions Herein, we integrated demographic information, eye tracking, and digital drawing clock tests to construct an MCI screening model that yielded superior classification performance. As a potential intelligent screening tool for MCI in the community, we aim to further build a multicenter external validation study to improve the model's generalizability.