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

TL;DR

This study investigates the feasibility of assessing cognitive impairment in MCI using a privacy-preserving, edge-computing distributed camera network in a real-world, group-based setting. By deploying 39 edge devices over a 1,700 m^2 space, it collects movement and social interaction cues during breaks, extracts a rich set of movement and social features, and trains multiple classifiers to distinguish high- versus low-functioning cohorts defined by MoCA scores. The findings show significant differences in several features and achieve up to 0.71 accuracy and 0.68 F1 in cohort-level classification, even without individual identifiers, highlighting the potential of longitudinal, low-cost passive monitoring to support dementia risk assessment. The work underscores practical implications for deployment in clinical or care settings and outlines future directions, including sensor fusion with Bluetooth and temporal deep learning for tracking longitudinal cognitive changes.

Abstract

INTRODUCTION: Mild cognitive impairment (MCI) is characterized by a decline in cognitive functions beyond typical age and education-related expectations. Since, MCI has been linked to reduced social interactions and increased aimless movements, we aimed to automate the capture of these behaviors to enhance longitudinal monitoring. METHODS: Using a privacy-preserving distributed camera network, we collected movement and social interaction data from groups of individuals with MCI undergoing therapy within a 1700$m^2$ space. We developed movement and social interaction features, which were then used to train a series of machine learning algorithms to distinguish between higher and lower cognitive functioning MCI groups. RESULTS: A Wilcoxon rank-sum test revealed statistically significant differences between high and low-functioning cohorts in features such as linear path length, walking speed, change in direction while walking, entropy of velocity and direction change, and number of group formations in the indoor space. Despite lacking individual identifiers to associate with specific levels of MCI, a machine learning approach using the most significant features provided a 71% accuracy. DISCUSSION: We provide evidence to show that a privacy-preserving low-cost camera network using edge computing framework has the potential to distinguish between different levels of cognitive impairment from the movements and social interactions captured during group activities.

Figures (5)

• Figure 1: The layout of the indoor space used in this study, spanning 1,700$m^2$, along with pictures from various functional areas within this space. Our study site has various regions to provide physical and cognitive training relating to activities in daily living for individuals with MCI. These areas include a gym, dining area, kitchen, lounge, activity area, tech bar, and staff zone. A picture of one of the cameras installed in the ceiling of the tech bar is also shown.
• Figure 2: Box plot illustrating the median, interquartile range, maximum and minimum of MoCA scores for each cohort. The MoCA scores for 66 subjects belonging to one of six cohorts are represented by the blue circles. Note that multiple individuals can have the same MoCA scores, resulting in overlapping blue circles, which appear as darker blue circles. The mean MoCA score for each cohort is displayed at the top of the plot. The threshold of 21 that is used to assign cohorts as either high functioning or low functioning is depicted by the dotted red line. The class that each cohort belongs to, i.e. high or low functioning, is denoted next to the cohort names.
• Figure 3: Proposed pipeline. A distributed camera network uses real-time pose estimation to collect privacy-preserving data from individuals in an indoor space. These keypoints are used to find the locations, orientations, and tracks of individuals in the indoor space kwon2023indoor. This is further used to identify and locate group formations hegde2024_group. Hand crafted features are extracted from the positions, orientations, tracks and group formation estimations. These features are used to classify a cohort as either a high- or low-functioning MCI cohort.
• Figure 4: Locations and coverage of the 39 edge computing camera systems deployed in the ceiling of the indoor space. The blue triangles depict the camera placement position and orientation. The cameras point in a direction away from the vertex, perpendicular to the short base, with a solid angle viewing region denoted by the shaded red regions. White areas are not covered by cameras, either by choice or infrastructural limitations.
• Figure 5: Feature importance analysis for the SVM models used for classification. The top figure shows the feature weights when using all features, the middle figure shows the feature weights when only using social interaction features, and the bottom figure shows the feature weights when only using movement features.