Premium
A comparison of classification methods for differentiating fronto‐temporal dementia from Alzheimer's disease using FDG‐PET imaging
Author(s) -
Higdon Roger,
Foster Norman L.,
Koeppe Robert A.,
DeCarli Charles S.,
Jagust William J.,
Clark Christopher M.,
Barbas Nancy R.,
Arnold Steven E.,
Turner R. Scott,
Heidebrink Judith L.,
Minoshima Satoshi
Publication year - 2004
Publication title -
statistics in medicine
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.996
H-Index - 183
eISSN - 1097-0258
pISSN - 0277-6715
DOI - 10.1002/sim.1719
Subject(s) - voxel , linear discriminant analysis , artificial intelligence , positron emission tomography , dementia , pattern recognition (psychology) , medical diagnosis , logistic regression , computer science , quadratic classifier , data set , nuclear medicine , medicine , machine learning , pathology , support vector machine , disease
Flurodeoxyglucose positron emission tomography (FDG‐PET) is being explored to determine its ability to differentiate between a diagnosis of Alzheimer's disease (AD) and fronto‐temporal dementia (FTD). We have examined statistical discrimination procedures to help achieve this purpose and compared the results to visual ratings of FDG‐PET images. The methods are applied to a data set of 48 subjects with autopsy confirmed diagnoses of AD or FTD (these subjects come from a multi‐centre collaborative study funded by the National Alzheimer's Coordinating Center). FDG‐PET images are composed of thousands of voxels (volume elements) so one is left with a situation where there are vastly more variables than subjects. Therefore, it is necessary to perform a data reduction before a statistical procedure can be applied. Approaches using both the entire image and summary statistics calculated on a number of volumes of interest (VOI) are examined. We performed the data reduction techniques of principal components analysis (PCA) and partial least‐squares (PLS) on the entire image and then used linear discriminant analysis (LDA), quadratic (QDA) or logistic regression (LR) to classify subjects as having AD or FTD. Some of these methods achieve diagnostic accuracy (as assessed by leave‐one‐out cross‐validation) that is similar to visual ratings by expert raters. Methods using PLS appear to be more successful. Averaging or using VOI data may also be helpful. Copyright © 2004 John Wiley & Sons, Ltd.