Leveraging Naturalistic Driving Digital Biomarkers for Early Mild Cognitive Impairment Detection: Deep Learning...

Area of Science:

• Neurology
• Digital Health
• Machine Learning

Background:

• Alzheimer disease and related dementias are a growing global health concern.
• Early detection of mild cognitive impairment (MCI) is crucial for timely intervention.
• Naturalistic driving behavior offers a promising, noninvasive digital biomarker for cognitive assessment.

Purpose of the Study:

• To develop deep learning strategies for early MCI detection using naturalistic driving data.
• To overcome limitations of prior research relying on controlled settings or simplified data.
• To establish driving behavior as a digital biomarker for cognitive decline.

Main Methods:

• Collected in-vehicle sensor data (GPS, accelerometer, gyroscope) from participants driving naturally.
• Segmented data into full trips and turning maneuvers for analysis.
• Compared three deep learning modeling strategies: single-view, feature-level fusion, and model-level late fusion.

Main Results:

• Models using full-trip data achieved 78% accuracy and 77% AUC, outperforming turn-only analysis.
• Late fusion of turn and trip data improved performance but did not surpass the full-trip baseline.
• Classification accuracy increased with more data, and trip-wise modeling better captured MCI's episodic nature.

Conclusions:

• Naturalistic driving behavior is a scalable, noninvasive method for early cognitive screening.
• Deep learning models analyzing full-trip driving data show potential for MCI detection.
• This framework supports real-world monitoring and digital health interventions for dementia prevention.