COMPARISON OF NON-UNIFORMITY-CORRECTION ALGORITHMS USING 
MULTIPLE REPEAT T1-WEIGHTED MRI SCANS OF A SINGLE SUBJECT

Six T1-weighted brain volumes of a normal volunteer were acquired over a six-month period on a Siemens Vision 1.5T MRI scanner using the 3DFLASH protocol (TR 35 msec, TE 6 msec, flip angle 45 degrees, one excitation, 165 x 220 mm FOV, 192 x 256 matrix, 100 slices, voxel dimensions 0.86 x 0.86 x 2.0 mm). Stripping (cortical isolation) masks were created for each volume as described above, and the raw volumes and their corresponding masks were aligned to a common coordinate system using the AIR 3.0 algorithm; to insure that data interpolation effects were similar for the six test volumes, for each volume an average transform to a Talairach reference volume was employed. After the non-uniformity-correction algorithms were run on the six aligned brain volumes, each corrected data volume was mean-normalized to the corresponding input volume to facilitate quantitative comparisons between algorithms.

    - Biased Brain Phantom Data (Figs. 1-4)
    - High-Resolution T1-weighted MRI Scans from Two Different Centers (Figs. 11-12)

    - Return to Introduction Page

Figure 5     Figure 6     Figure 7     Figure 8     Figure 9     Figure 10

Figure 9
(A) Scatter plot of the first two canonical variates (CV1, CV2) from a canonical variables analysis of the six repeat scans before and after non-uniformity correction. spm was excluded from the analysis, because if it was included, 98% of the variance in the data decomposition was accounted for by difference between spm and the other five methods along the CV1* dimension (inset). Note that the eq cluster (closed triangles) is separated from the other four corrected volume clusters and the uncorrected volumes along the CV1 axis, while the CV2 axis separates the hum and cma clusters (closed and open diamonds) from the other three corrected volume sets and the uncorrected volumes. (B) The canonical eigenimage corresponding to CV1. Note the central hypodensity and the prominence of structural detail. (C) The canonical eigenimage corresponding to CV2.  Non-uniformity is most marked along the axial direction. Note that (B) and (C) are scaled differently.
 
 
 

A
 
B - Canonical eigenimage corresponding to CV1
 
 
C - Canonical eigenimage corresponding to CV2
 

© 2000 by the International Neuroimaging Consortium with funding from the 
Human Brain Project and by the Regents of the University of Minnesota. 
Contact: webmaster@pet.med.va.gov		 The University of Minnesota is an equal opportunity educator and employer.