JMIR Publications

ABSTRACT

Background:

The prompt and accurate identification of Mild Cognitive Impairment (MCI) is crucial for preventing its progression into more severe neurodegenerative diseases. However, current diagnostic solutions, such as biomarkers and cognitive screening tests, prove costly, time-consuming, and invasive, hindering patient compliance and the accessibility of these tests. Therefore, exploring a more cost-effective, efficient, and non-invasive method to aid clinicians in detecting MCI is necessary.

Objective:

This study aims to develop an ensemble learning framework that adaptively integrates multimodal physiological data collected from wearable wristband and digital cognitive metrics recorded on tablet, thereby improving the accuracy and practicality of MCI detection.

Methods:

In this study, we recruited 843 participants aged 60 and above from the First Hospital of Shanxi Medical University, who were randomly divided into a development dataset (674 participants) and an internal test dataset (169 participants) at a 4:1 ratio. Additionally, 226 older adults were recruited from three external centers to form an external test dataset. We measured their physiological signals (including electrodermal activity and photoplethysmography) and digital cognitive parameters (such as reaction time and test scores) using the clinically certified Empatica 4 wristband and a tablet cognitive screening tool. The data were then preprocessed, and features in the time, frequency, and nonlinear domains were extracted from the individual physiological signals. To mitigate the challenges associated with dimensionality and model complexity arising from the multidimensional features (comprising physiological signals and cognitive data), we propose a dynamic adaptive feature selection optimization algorithm to generate a subset of features that contributed the most to classification performance. Finally, by optimizing the combination of base learners, we improved the accuracy and efficiency of the classification model.

Results:

The experimental results indicate that the proposed MCI detection framework achieved classification accuracies of 88.4%, 85.5%, and 84.5% on the development, internal test, and external test datasets, respectively. The area under the curve for the binary classification task was 0.945 (95% CI 0.903-0.986), 0.912 (95% CI 0.859-0.965), and 0.904 (95% CI 0.846-0.962) on these datasets. Furthermore, a statistical analysis of feature subsets during the iterative modeling process revealed that the decay time of skin conductance response (SCR), the percentage of continuous normal-to-normal intervals exceeding 50 milliseconds (PNN50), the ratio of low-frequency to high-frequency (LF/HF) components in heart rate variability, and cognitive time features emerged as the most prevalent and effective indicators. Specifically, compared to healthy individuals, MCI patients exhibited a longer SCR decay time during cognitive testing (p<0.001), lower PNN50 (p<0.001), and higher LF/HF (p<0.001), accompanied by greater variability. Similarly, MCI patients took longer to complete cognitive tests than healthy individuals (p<0.001).

Conclusions:

The developed MCI detection framework has demonstrated significant potential as an effective and efficient tool for detecting MCI. It establishes a new benchmark for non-invasive, cost-effective early MCI detection that can be integrated into routine wearable and tablet-based assessments. This approach not only enhances accessibility but also provides a feasible solution for autonomous, at-home monitoring and early detection, making it a valuable tool in the ongoing fight against neurodegenerative diseases.