Optimization of Neuropsychological Scores at the Baseline Visit Using Evolutionary Technique

Abstract

The neuropsychological battery of scores, are the measures of cognitive domains of human brain, that are considered as important features to distinguish healthy subjects from the subjects, suffering from Mild Cognitive Impairment (MCI). The instances of about 5542, with four time visits are separated from the total collected instances of the National Alzheimer’s Coordinating Center (NACC) database. The analysis of the selected data shows that the large number of subjects is identified for 66–75 and 76–85 age groups. The Genetic Algorithms (GA) applied on the neuropsychological scores at the baseline visit, selects the best subset of scores required for the clinical diagnosis, and these scores are evaluated by the logistic regression model using Area Under Curve (AUC), accuracy and Mean Squared Error (MSE) as the metric. Simulations result show that a highest classification accuracy of 0.9427, AUC of 0.9713 and less error rate of 0.041 is achieved for the combination of four neuropsychological scores Global Staging of Clinical Dementia Rating (CDRGLOB), Geriatric Depression Scale (GDS), Logical Memory Delayed (MEMUNITS), Digit Span Forward Length (DIGIFLEN). These scores are predominantly selected by the GA across many runs and thus have significant role for screening MCI subjects at the baseline visit.