Subject population:
Two subjects (age 38 – subject 1 – and age 49 – subject 2 –,
males) underwent head MRI on a tilting 0.25 T system (GScan Brio,
Esaote,
Italy) and on a 1.5 T scanner (Magnetom Avanto,
Siemens,
Germany).
Data acquisition:
An isotropic (resolution 1 × 1 × 1 mm3) high-resolution T1-weighted 3D image,
using a magnetization-prepared rapid acquisition with gradient echo (MPRAGE) sequence,
was acquired on both scanners in the supine position.
The sequence parameters for 0.25 T were the following: TR=12 ms,
TE=6 ms,
200 axial slices,
in-plane matrix size= 256 × 256.
The sequence parameters for 1.5 T were the following: TR=1900 ms,
TE=3.37 ms,
TI=1100 ms,
176 axial slices,
in-plane matrix size= 192 × 256.
For the 49 years old subject,
the same sequence was also acquired during WB on the 0.25 T system and twice,
after repositioning,
on the 1.5 T scanner.
The Speed Up technique,
based on the mathematical theory of compressed sensing [2],
was used to optimize acquisition time and SNR on the 0.25 T system.
Data analysis:
In order to compare the quality of the 3D T1 images obtained from the different scanners and between supine and WB positions,
we evaluated: 1) the volumes of different segmented tissues; 2) the images obtained from the same subjects with the different scanners (or positions),
coregistering them to the 1.5 T images; 3) the SNR of the whole brain; 4) the contrast of the different tissues.
All the MRI data were processed with FMRIB’s Software Library (FSL,
http://www.fmrib.ox.ac.uk/fsl).
The non-brain tissue was removed from the MPRAGE images,
using the Brain Extraction Toolbox (BET) [3].
Then,
each subject's MPRAGE image was segmented into gray matter (GM),
peripheral GM (pGM),
white matter (WM),
and ventricular cerebrospinal fluid (vCSF) with SIENAX [4].
In addition,
right/left thalamus,
caudate,
putamen,
pallidum,
hippocampus,
amygdala,
and nucleus accumbens were segmented using FIRST [5] on the betted data.
The volumes of the GM,
pGM,
WM,
vCSF,
and subcortical GM (SGM) structures were computed.
Brain volume was estimated as the sum of WM and GM volumes.
The rigid transformation for coregistering the 0.25 T 3D T1 on the 1.5 T MPRAGE was estimated and applied to the GM mask for a visual comparison of the GM segmentations obtained with the different scanners.
The SNR was computed for the whole brain from the following equation,
as suggested in our previous work [6]:
SNR=0.655*S/N
Where:
- N is the noise,
estimated for each raw image as the standard deviation of the signal intensity extracted from six areas,
carefully drawn in the background (air) outside the brain.
- S is the mean signal in the brain.
The brain mask was obtained as the sum of the eroded GM and WM masks.
The GM and WM tissues were eroded in order to increase the confidence of extracting the signal from the brain tissue.
The contrasts between the different couples of tissues (GM,
WM,
vCSF) were computed using the following equation:
Contrast(AvsB)=(A-B)/(A+B)
where A and B are the couple of tissue of interest.
The contrasts between WM and GM; WM and CSF; GM and CSF were evaluated and compared between the two scanners.
For each subject,
the volumes,
SNR and Contrasts obtained with the two scanners,
were compared computing the difference of the corresponding values,
normalized by their average.
The scan-rescan repeatability was estimated at 1.5 T,
computing the difference of corresponding volumes,
normalized by their average.
These values were used as reference when evaluating the differences between scanners.