Premium
IC–P–065: Mapping of neuronatomical changes associated with ApoE genotype in Alzheimer's disease
Author(s) -
Ewers Michael,
Hampel Harald,
Born Christine,
Schoenberg Stefan O.,
Moeller Hans-Juergen,
Teipel Stefan J.
Publication year - 2006
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.2006.05.2270
Subject(s) - apolipoprotein e , grey matter , parahippocampal gyrus , voxel based morphometry , voxel , hippocampus , psychology , amygdala , alzheimer's disease , medicine , neuroscience , pathology , temporal lobe , magnetic resonance imaging , white matter , disease , radiology , epilepsy
pattern, and a statistical pattern recognition method can be used to map different patterns to different group membership. Objective(s): To investigate if machine learning methods can classify subjects between Alzheimer’s disease and healthy control (HC) subjects. Methods: We applied the Support Vector Machine (SVM) algorithm to perform classification of group membership using resting glucose metabolism images obtained using positron emission tomography (PET). There were 17 AD and 12 age-matched HC. The average (standard deviation) age was 67.8 (10.1) and 66.8 (9.8) years for the HC and AD patients, respectively. The average Mini Mental State Exam score was 29.4 (0.6) and 11 (6.7) for the HC and AD patients, respectively. Values of the resting glucose metabolism were obtained using a regions of interest (ROI) in the visual cortex, lateral temporal cortex, medial temporal lobe, inferior temporal lobe, ventral and dorso-lateral frontal lobe, anterior cingulate, and posterior cingulate. Results: We found statistically significant differences in the temporal lobe, inferior parietal region, posterior cingulate and medial and frontal lobe areas. The analysis suing the SVM had two phases: during the training, the classifier algorithm finds the set of regions by which the two groups can be best distinguished from each other. In the test phase, given the FDG-PET values from a new subject the classifier predicts the subject’s group membership. Conclusions: Our approach leads to two outputs: the classification of group membership, and the discriminating regions upon which the classification is based on. This method can be expanded to discriminate among 3 or more groups. The method is very promising as a tool to assist in the classification of subjects to patient populations.