Feasibility of assessing cognitive impairment via distributed camera network and privacy-preserving edge computing

Mild cognitive impairment (MCI) involves cognitive decline beyond normal age and education expectations. It correlates with decreased socialization and increased aimless motion. We aim to automate detection of these behaviors for improved longitudinal monitoring.

We used a privacy-preserving distributed camera network to collect data from MCI patients in an indoor space. Movement and social interaction features were developed using this data to train machine learning algorithms to differentiate between higher and lower cognitive functioning MCI groups.

A Wilcoxon rank-sum test showed significant differences between high- and low-functioning cohorts in the movement and social interaction features. Despite the absence of data linking each person's identity to their specific level of cognitive decline, a machine learning model using key features achieved 71% accuracy.

We show that an edge computing-based privacy-preserving camera network can differentiate between levels of cognitive impairment based on movements and social interactions during group activities.

Movement and social interaction features showed significant differences in high- and low-functioning cohorts.Significant features included linear path lengths, walking speed, direction change and velocity entropies, and number of group formations, among others.Differences were observed despite the presence of healthy individuals and the lack of individual identifiers.Data were collected using a 39-camera privacy-preserving edge computing network covering a 1700-m² indoor space.