Deep learning for early detection of mild cognitive impairment using smart home ambient sensor data

Area of Science:

• Gerontology
• Artificial Intelligence
• Biomedical Informatics

Background:

• Early detection of mild cognitive impairment (MCI) is crucial for timely intervention.
• Traditional machine learning methods struggle with complex temporal patterns in sensor data.
• Smart home ambient sensors offer continuous, unobtrusive monitoring of daily behaviors.

Purpose of the Study:

• To develop and evaluate a deep learning model for classifying older adults into healthy and MCI categories using smart home sensor data.
• To compare the performance of a Temporal Convolutional Network with Contrastive Pretraining (TCN-CL) against traditional machine learning classifiers.
• To assess the potential of TCN-CL in improving diagnostic accuracy for cognitive decline in diverse home environments.

Main Methods:

• 137 community-dwelling older adults (76 healthy, 61 MCI) were monitored for 30 days.
• 34 digital markers (sleep, activity, regularity) were extracted from ambient smart home sensors.
• A TCN-CL model was trained and compared with logistic regression and decision tree classifiers.

Main Results:

• TCN-CL significantly outperformed baseline models in predicting cognitive diagnoses.
• TCN-CL achieved 85% accuracy, 0.77 sensitivity, 0.92 specificity, and a 0.71 Matthews correlation coefficient.
• Traditional models like logistic regression and decision tree showed poor performance.

Conclusions:

• Pretrained TCN-CL provides a robust method for early cognitive decline detection via smart home sensors.
• Continuous behavioral monitoring using ambient sensors can enhance clinical decision-making.
• This approach facilitates accurate and timely diagnosis of MCI.