Alzheimer’s Disease Neuroimaging Initiative

Data and Samples
>
ADNI Data
>
Neuroimaging
>
MRI
>
MRI Pre-processing

MRI Pre-processing

The ADIR Lab provided intensity normalized and gradient un-warped T1 image volumes for all ADNI1 and many ADNI2 scans. As MR vendors offered these corrections online as part of the product, ADNI stopped performing its own preprocessing and instead employed the preprocessing performed by vendor product on the scanner. Consequently, no offline T1 image preprocessing is necessary for ADNI3 and ADNI4 images. Siemens scanners may save two variants of 3D scans with and without gradient linearity correction applied (gradwarp).

ADIR Lab QC will select the variant with gradwarp, but if users need to disambiguate these cases, the gradwarp corrected version
of the image will indicate “DIS3D” in the DICOM header field Image Type (0008,0008) for the series.

All 3D T1 images from ADNIs 3 and 4 include an on-scanner non-uniformity correction, this may not fully eliminate non-uniformity. Many volumetric analysis pipelines now include bias field correction, however users not using such a pipeline and performing analysis which may be influenced by non-uniform intensities within the image may wish to apply an additional bias field correction such as N4 (Tustison et al. IEEE TMI 2010, 29(6)).

ADNI1 and ADNI2/GO Data Pre-Processing for 3D T1 Image Volumes

Each MPRAGE image in the database at LONI is linked with related image files, which have undergone specific image pre-processing correction steps. These corrections are as follows:

Gradwrap Correction

Gradwarp is a system-specific correction of image geometry distortion due to gradient non-linearity. The degree to which images are distorted due to gradient non-linearity varies with each specific gradient model. We anticipate that most users will prefer to use images which have been corrected for gradient non-linearity distortion in analyses.

B1 Non-Uniformity Correction

This correction procedure employs the B1 calibration scans noted in the protocol above to correct the image intensity non-uniformity that results when RF transmission is performed with a more uniform body coil while reception is performed with a less uniform head coil.

N3 Non-Uniformity Correction

N3 is a histogram peak sharpening algorithm that is applied to all images. It is applied after gradwarp and after B1 correction for systems on which these two correction steps are performed. N3 will reduce intensity non-uniformity due to the wave or the dielectric effect at 3T. 1.5T ADNI1 and some ADNI2 3D T1 scans underwent N3 processing to reduce residual intensity non-uniformity.

The need to perform the image pre-processing corrections outlined above varies with manufacturer and system RF coil configuration. Philips systems were equipped with B1 correction as product at the time ADNI began. In addition, Phillips gradient systems tend to be linear. Therefore, no gradwarped and no B1 corrected pre-processed files are generated for images acquired on Phillips systems. The files available by manufacturer will be:

Phillips Systems:	GE and Siemens systems with transmit-receive head RF coils:	GE and Siemens systems with receive-only head RF coils:
Raw DICOM N3 corrected	Raw DICOM Grad warped Grad wrap plus N3	Raw DICOM Grad warped Grad wrap plus N3

As noted above, it is anticipated that nearly every user will prefer to employ scans which have undergone gradwarp correction in analyses. Users who have developed their own set of tools for image intensity corrections may wish to simply use the gradwarped files. However, it is anticipated that most users will want to use the fully pre-processed files. These are most easily identified as files that contain N3 in the identifier. Note that these corrections are applied only to T1-weighted (not FSE) images, and as outlined below, only to the one T1-weighted volume associated with each time point that has been designated as best by the ADNI QC team.

Phantom-Based Scaling Measures

In addition to the corrections outlined above, for ADNI1 and early ADNI2/GO, phantom-based measures of spatial scaling are associated with each MPRAGE image volume in an accompanying XML file. A version of the image with these spatial scale factors applied is provided.

Recall that each ADNI1 and early ADNI2/GO human scans was followed immediately by an acquisition with the ADNI phantom. Absolute scaling along each of the cardinal axes (x, y, z) is measured with the phantom. These phantom-based measurements were used to retrospectively scale the accompanying human MPRAGE image. In the limit that the image matrix is aligned with the cardinal axes, this amounts to adjusting the voxel size.

For images for which this does not hold, application of the scale factors is slightly more complicated, as scaling along one axis in the magnet will be mixed into the other two dimensions by the oblique rotation.