Premium
IC‐P‐123: Gray matter correlates of Alzheimer's disease risk as quantified with latent information in semantic fluency lists
Author(s) -
Clark David,
Hwang Kristy,
McLaughlin Paula,
Woo Ellen,
Babakchanian Sona,
Apostolova Liana
Publication year - 2013
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1016/j.jalz.2013.05.120
Subject(s) - verbal fluency test , raw score , psychology , semantic memory , artificial intelligence , dementia , audiology , neuropsychology , cognition , cognitive psychology , statistics , computer science , disease , medicine , pathology , mathematics , neuroscience , raw data
Background: Semantic fluency tasks are valuable tools for diagnosing dementia. The order ofwords in fluency lists reflects the organization of semantic memory and may be distorted in the setting of brain disease. Random forest classifiers trained to predict conversion to Alzheimer disease (AD) using latent information from semantic fluency lists perform better than classifiers trained using only verbal fluency raw scores. The current research makes use of risk estimations from the latent information classifiers in a new sample of elderly individuals, evaluating the relationship of estimated risk with gray matter thickness.Methods: One hundred forty-three subjects (48 controls, 95 subjects with mild cognitive impairment) underwent two of the five semantic fluency tasks on which the classifiers in Clark et al. were based (’animals’ and ’vegetables’). Word lists from these tasks were entered into classifiers to generate estimates of conversion risk. All participants underwent high-resolution MRI scans that were processed by previously described methods for detailed analysis of gray matter thickness. Estimates of the risk for conversion toADwere entered as a predictor variable in amultiple linear regressionmodelwith graymatter thickness as the dependent variable and nuisance covariates of sex, education, and age. P-values from the regression model were plotted on a three-dimensional model of the cerebral cortex. Correction formultiple comparisonswas undertakenwith a permutation analysis.Results: Estimated risk of conversion to AD (as estimated by the latent information classifiers) was associated with broad regions of gray matter thinning in the lateral and mesial parietal lobes, posterior temporal lobes, and mesial frontal lobes of both hemispheres. Conclusions: Estimates of conversion risk generated by latent information classifiers are associated with gray matter thinning in regions known to become abnormal in patients with AD. These results support the utility of latent information classifiers for predicting functional decline in individuals at risk for AD.