In this study the association between CSF and structural imaging based early biomarker of Alzheimer Diasease (AD) was investigated in amnestic Mild Cognitive Impairment (aMCI) patients. Voxel based morphometry and partial least square correlation were used to analyze differences in local and whole brain GM profiles (covariance) between two groups of aMCI patients divided in two groups based on CSF amyloid (Aβ42) levels. Both voxel-wise and structural covariance GM differentiated Aβ positive (prodromal AD) from Aβ negative patients. These results indicate that GM measures may provide a sensitive metric for tracking AD progression.
Participants: 144 aMCI patients enrolled in 13 European clinical sites in
the IMI WP5 PharmaCog (also referred to as the European ADNI study). They
were classified (high risk, n=81) or not (low risk, n=63) as Prodromal AD depending
on whether their Aβ42 CSF level were low (Aβ positive) or high (Aβ negative)2 .The two groups were age and gender matched3.
Image acquisition: T1-weighted images were acquired on 3.0 T scanners of different MRI system vendors and models using either a MPRAGE (Siemens and Philips) or a IR-SPGR (GE) sequence with the following acquisition parameters: 3D sagittal acquisition, square FOV=256 mm, 1 × 1 × 1 mm3, TR/TI=2300/900 ms, flip angle = 9°, no fat suppression, full k-space, no averages4.
Image preprocessing: The CAT tool in SPM 12 was used for segmentation/normalization. Images were first segmented into GM, white matter, and cerebrospinal fluid. Quality control procedures were conducted prior to subsequent steps to check for inhomogeneities and general quality of the segmentation and no images were discarded. Subsequently GM maps were aligned to a standard DARTEL template in MNI space, modulated and smoothed to 8 mm FWHM.
Voxel based morphometry (VBM): A two independent samples T-test was run to assess voxel-wise differences in GM between the two groups of amnestic MCI patients (high risk and low risk). Age and total intracranial volume were included as covariates. Only voxels with absolute value above 0.2 were selected to be included in the analysis. A voxel-level threshold of p<0.05 family-wise error corrected (FWE) was considered for statistical significance.
Structural Covariance Analysis: In order to detect whole brain structural covariance differences between high and low risk patients a mean centered partial least square correlation (PLS) was performed (PLSgui version 6.13 on MATLAB, 2013): this methods allows to first construct the structural covariance matrix of GM for the whole dataset and then to derive individual participant’s score (CovSc) reflecting how much of the covariance pattern is present in each brain image. Therefore we calculated and compared using a two sample independent test the CovSc between our two groups of aMCI patients. Being a multivariate methods PLS offers the advantage of testing all brain voxels together and does not require multiple comparison correction.