Screening for Early-Stage Alzheimer's Disease Using Optimized Feature Sets and Machine Learning.

Area of Science:

• Neurology
• Gerontology
• Cognitive Science

Background:

• Early-stage Alzheimer's disease detection is hindered by the lack of sensitive, practical cognitive assessments for mild impairment.
• This contributes to a high rate of undetected dementia in clinical settings.

Purpose of the Study:

• To identify optimal cognitive assessment features for detecting mild impairment and improving routine screening.
• To compare the effectiveness of two-class (impaired/non-impaired) versus three-class (CDR 0, 0.5, 1) classification for dementia staging.

Main Methods:

• Supervised feature selection was used to identify cognitive measurements for impairment (CDR 0.5+).
• Random forest classifiers and stochastic cross-validation predicted impairment, with results analyzed using general linear models.

Main Results:

• A two-class classification strategy significantly improved sensitivity and negative predictive values compared to a three-class approach.
• Four features (Logical Memory, trail-making, memory questions) achieved 94.53% sensitivity in ~15 minutes.
• Adding four more features increased sensitivity to 95.18%.

Conclusions:

• A concise set of four cognitive features shows high detection rates for mild cognitive impairment (CDR 0.5+).
• These features offer a practical foundation for developing effective cognitive impairment screening protocols.