One-class classification with confound control for cognitive screening in older adults using gait, fingertapping,...

Area of Science:

• Neuroscience
• Machine Learning
• Gerontology

Background:

• Early detection of cognitive impairment is critical for dementia intervention.
• Current screening tools are not optimal for widespread population use.
• Developing accessible screening methods is a public health priority.

Purpose of the Study:

• To explore the efficacy of one-class classification with motor-cognitive tasks for early cognitive impairment detection.
• To compare the performance of one-class classification against two-class classification and standard screening tests.
• To investigate the optimal method for combining data from various tasks.

Main Methods:

• Data collected on gait, fingertapping, cognitive, and dual tasks from older adults (mild cognitive impairment and healthy controls).
• One-class classification modeled healthy control behavior; deviations flagged as abnormal.
• Confound regression integrated; individual tasks, early/late fusion evaluated.

Main Results:

• One-class classification demonstrated higher predictive accuracy for mild cognitive impairment.
• Gait features were most effective for one-class classification.
• Combined models outperformed combined features; one-class majority voting achieved 87.5% sensitivity and 75.7% specificity.

Conclusions:

• One-class classification with confound control effectively detects abnormal motor-cognitive patterns for early cognitive impairment detection.
• This method shows potential as an alternative to standard cognitive screening.
• Further validation in diverse clinical settings is warranted.