A Stable and Scalable Digital Composite Neurocognitive Test for Early Dementia Screening Based on Machine Learning:...

Area of Science:

• Neuroscience
• Artificial Intelligence
• Public Health

Background:

• Dementia presents a significant public health challenge.
• Mild cognitive impairment (MCI) is a critical transitional phase preceding dementia.
• Early MCI detection is vital for dementia prevention strategies.

Purpose of the Study:

• To develop and validate a stable, scalable cognitive test panel for early MCI and dementia detection.
• Utilize machine learning (ML) on the Chinese Neuropsychological Consensus Battery (CNCB).
• Assess performance in the Chinese Neuropsychological Normative Project (CN-NORM) and ADNI-3 cohorts.

Main Methods:

• Recruited 871 participants across normal, MCI, and dementia groups in the CN-NORM study.
• Employed four ML algorithms (SelectKBest, logistic regression, logit, stepwise elimination) for feature selection.
• Validated models using receiver operating characteristic curves, AUC, and cross-validation in CN-NORM and ADNI-3 cohorts.

Main Results:

• The Hopkins Verbal Learning Test-5 minutes Recall showed the highest discrimination (AUC=0.80) between MCI and normal cognition.
• A scalable model (Model 5: HVLT-5min Recall and Trail Making Test-B) demonstrated superior discrimination to existing tests.
• Model 5 achieved high sensitivity (0.82-0.83) and validated robustly in the ADNI-3 cohort (AUC=0.81).

Conclusions:

• Developed a stable, scalable ML-based neurocognitive test for differentiating cognitive impairment stages.
• The composite test serves as a feasible digital biomarker for large-scale screening.
• This tool has potential applications in cognitive screening and intervention studies.