1Neuroimaging Laboratory, Santa Lucia Foundation, Rome, Italy, 2Clinical Imaging Sciences Centre, Brighton and Sussex Medical School, University of Sussex, Brighton, United Kingdom
The current study is an application of
nested support vector machine (SVM) to distinguish healthy subjects
and patients with Alzheimer’s disease using very few features
coming from structural (T1) and diffusion (DWI) MR. After having
segmented the T1 images in GM, WM and CSF, mean values of
fractional_anisotropy, mean_diffusivity, radial_diffusivity and
axial_diffusivity were computed in GM and WM; volume of GM and WM as
percentage of total_intracranial_volume were also assessed.
Therefore we computed 1023 different SVMs, one for each possible
combination of the 10 features. Surprisingly, the WM diffusion measures resulted to be the most specific of dementia status.