Diffusion tensor based Alzheimer image analysis using region specific volume features and random forest classifier

In this work a method to utilize volume information as an independent feature for diagnosis of Alzheimer's disease (AD) using Diffusion Tensor Images (DTI) is proposed. For this study the DTI were obtained from ADNI, an international repository for current Alzheimer's study. Equal number of AD and normal control subjects were used for the study. The volume of six regions namely, Fornix/Stria Terminalis Left, Fornix/Stria Terminalis Right, Fornix, Corpus Callosum, Cerebral Peduncle and Anterior Corona Radiata, reported to be prominently responsible for AD were extracted. Volume features corresponding to the above said feature set was used to classify AD and controls using random forest classifier. The data was also used for classification with significant components extracted using principle component analysis. The classification accuracy of 71.4% was achieved for full dataset which further improved to 85.7% on application of principle components as the feature. An enhancement in recall was observed which increased from 71.4% for full dataset to 88.9% for principle components. The precision was observed to be 88.9%. The results demonstrate that volume can be used as a feature for differentiating AD from normal controls. The volume feature can be used independently for initial screening tests. This can be used for automated identification of AD using DTI.