A Genetic Algorithm for the selection of structural MRI features for classification of Mild Cognitive Impairment and Alzheimer's Disease

摘要

This work investigates the problem of feature selectionudin neuroimaging features from structural MRI brain imagesudfor the classification of subjects as healthy controls, sufferingudfrom Mild Cognitive Impairment or Alzheimer’s Disease. A GeneticudAlgorithm wrapper method for feature selection is adoptedudin conjunction with a Support Vector Machine classifier. In veryudlarge feature sets, feature selection is found to be redundant asudthe accuracy is often worsened when compared to an SupportudVector Machine with no feature selection. However, when justudthe hippocampal subfields are used, feature selection shows audsignificant improvement of the classification accuracy. Three-classudSupport Vector Machines and two-class Support VectorudMachines combined with weighted voting are also compared withudthe former and found more useful. The highest accuracy achievedudat classifying the test data was 65.5% using a genetic algorithmudfor feature selection with a three-class Support Vector Machineudclassifier.