Genetic algorithm-based PCA eigenvector selection and weighting for automated identification of dementia using FDG-PET imaging

Parametric FDG-PET data offer the potential for an automated identification of the different dementia syndromes. Principal component analysis (PCA) can be used for feature extraction in FDG-PET. However, standard PCA is not always successful in delineating the features that have the best discrimination ability. We report a genetic algorithm-based method to identify an optimal combination of eigenvectors so that the resultant features are capable of successfully separating patients with suspected Alzheimer's disease and frontotemporal dementia from normal controls. We compared our approach with standard PCA on a set of 210 clinical cases and improved the performance in separating the dementia types with an accuracy of 90.0% and a Kappa statistic of 0.849. There was very good agreement between the automated technique and the diagnosis given by clinicians.