Toward assessing functional decline in mild cognitive impairment using wearable sensors and explainable machine...

Area of Science:

• Gerontology and Cognitive Science
• Biomedical Engineering
• Machine Learning in Healthcare

Background:

• Mild cognitive impairment (MCI) increases fall risk and impacts independence in older adults.
• Progression to dementia is common but often detected late.
• Current assessments of instrumental activities of daily living (IADLs) lack real-world ecological validity.

Purpose of the Study:

• To develop and validate methods for detecting IADL subtasks using wearable sensors.
• To establish a foundation for longitudinal monitoring of cognitive and functional health.
• To identify digital biomarkers for early detection of functional decline in older adults.

Main Methods:

• An ecologically valid grocery shopping task was designed for 26 older adults (12 with MCI, 14 without).
• Participants were instrumented with inertial sensors to capture movement data.
• An interpretable deep learning framework was used for subtask detection, with SHAP for analysis.

Main Results:

• The deep learning framework accurately detected broad movement categories (e.g., Walk, Turn) but struggled with granular subtasks.
• Detection errors were linked to overlapping motion signatures between similar actions.
• SHAP analysis indicated orientation angles (yaw, roll) were key classification features.

Conclusions:

• Differences in feature weighting between cognitively unimpaired and MCI groups suggest potential digital biomarkers.
• Population-specific motor signatures may indicate early functional decline.
• This approach offers a foundation for objective, longitudinal monitoring of cognitive and functional status.