Automated classification of positron emission tomography images with mild cognitive impairment based on voxel of interest and neuropsychological test results

Alzheimer´s disease (AD) is the most common form of dementia with the unknown pathogenesis and pathologies. It brings about serious social problems. As predementia AD, mild cognitive impairment (MCI) subjects are usually overlooked because of the cryptic features of the occurrence and development of the disease. To detect MCI subjects from healthy controls as early and accurately as possible is of great importance and urgency to delay or prevent the onset of AD. In the paper, we propose a novel systematic method combining voxel of interest in positron emission tomography (PET) images and neu-ropsychological test results for automated classification of MCI subjects versus healthy controls (HC) in the Alzheimer´s Disease Neuroimaging Initiative (ADNI) database. The method includes four steps: pre-processing, extracting independent components, selecting voxels of interest, and classifying using a support vector machine classifier. PET images were obtained from ADNI database including 91 HC and 105 MCI patients with baseline diagnosis of MCI. As a result, we achieved good discrimination between MCI patients and HC with the averaged classification accuracy of 95.58%, sensitivity of 94.33%, and specificity of 97.04%. The experimental results show that the proposed method can successfully distinguish MCI from HC, and it is able to obtain higher classification accuracy of MCI versus HC than using only independent components or neuropsychological test results.