Premium
P3‐251: Automated semantic indices are related to functional brain networks in MCI and Alzheimer's disease
Author(s) -
Hemmy Laura,
Lim Kelvin,
Pakhomov Serguei
Publication year - 2012
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.2012.05.1474
Subject(s) - verbal fluency test , semantic similarity , neuropsychology , cluster analysis , computer science , similarity (geometry) , artificial intelligence , semantic memory , wordnet , natural language processing , task (project management) , neuropsychological assessment , psychology , cognitive psychology , cognition , neuroscience , management , economics , image (mathematics)
Background: Semantic verbal fluency (SVF) often shows early and disproportionate decline in AD relative to other language, attention, and executive abilities. Successful performance on SVF depends on the ability to organize conceptual information into related clusters and efficiently access these clusters. Current methods for clustering and switching assessment are labor-intensive and subjective. We developed an automated computational linguistic approach to quantify the semantic content of SVF responses. Methods: Neuropsychological and resting state fMRI data were obtained from the work-up of 52 patients presenting to the Minneapolis VAMC GRECCMemory Loss Clinic. Participants included had a clinical diagnosis ofMCI or AD, a completedMRI protocol with good quality data, and neuropsychological evaluation including the SVF task (animals). Imaging data were collected on a Philips 1.5T system at the Minneapolis VAMC. Semantic indices based on pairs of words on the SVF task were quantified in two ways: based on the length of their hierarchical relations inWordNet, an electronic lexical database of English (similarity), or calculated using a computerized algorithm based on a variant of principal components analysis (relatedness). Mean cumulative and sequential indices were produced for eachmethod: cumulative similarity and relatedness were computed between all possible pairs of words produced regardless of order; and sequential similarity and relatedness were computed only between pairs of adjacent words. Higher scores reflect larger clusters and reduced switching.Results: Several resting state fMRI network measures were related to the four automated semantic fluency indices. Nodal diversity, local efficiency, and the mean clustering coefficient, were all significantly correlated with cumulative and sequential measures of semantic similarity and relatedness. Pearson r values ranged from .323-.407 with corresponding p-values of .012-.003. All correlations survived multiple comparison correction. The traditional SVF score was not significantly related to imaging indices. Conclusions: We found that computational linguistic measurements of similarity and relatedness were significantly related to network measures obtained from resting state fMRI. These results suggest automated assessment of SVF has correlates with brain function in MCI and AD, and may outperform the traditional SVF score. This approach provides an easy way to standardize clustering and switching assessment without adding burden.