Image Processing Reference
In-Depth Information
A total of 876 scans from 340 children from the publicly available NIH-funded
MRI Study of Normal Brain Development [6] (termed here the NIH pediatric data-
base or NIHPD group; 179 female and 161 male; mean age at first scan 11.04 years,
range 4.88-19.76 years) were used as the normal growth reference. Participants in the
NIHPD study were scanned serially at two or three time points, with approximately
24 months between scans.
MRI Protocol
Magnetic resonance images of the MS and NC subjects were obtained using a single
GE 1.5T TwinSpeed Excite 12.0 scanner at the Hospital for Sick Children in Toronto,
Canada. The MRI protocol included a whole brain, 3D T1-weighted (T1w) RF-
spoiled gradient-recalled echo sequence (1.5-mm thick sagittal partitions, TR = 22
ms, TE = 8 ms, excitation pulse angle = 30◦, FOV = 250 mm, 256×256 matrix, voxel
size = 0.98×0.98×1.5 mm). To help in lesion segmentation, we acquired a 2D multi-
slice dual-echo PDw/T2w fast spin-echo sequence (2-mm thick slices, no gap, echo
train length = 8, TR = 3500 ms, TE1/TE2 [effective] = 15/63 ms).
Scans of the NIHPD controls were obtained at six study centers. The MRI protocol
included a whole brain, 3D T1w RF-spoiled gradient echo sequence (1-mm thick sagit-
tal partitions, TR = 22-25 ms, TE = 10-11 ms, excitation pulse angle = 30◦, FOV = 160-
180 mm). Additional acquisition and subject details are found in Evans et al.[6].
2.2
Methods
For each patient, the analysis method consists of 4 steps: (1) brain and thalamus
volume calculation for all time-points, (2) brain and normalized thalamus individual
curve model creation (3) brain and normalized thalamus volume and growth rate es-
timation from the model at the first attack (4) comparison with age and gender-
matched normal controls (NIHPD group).
We didn't compare directly the HSC patients with the subjects from the NC group
acquired on the same scanner. We compared the patients to the subjects of the NIHPD
group that had more normal controls subjects, more scans per subject and a wide age
range. However, the HSC NC group was essential to ensure that our scanner was
comparable to images acquired from the NIHPD study. To study a potential scanner-
related influence, we applied the same 4 analysis steps at each subject of the NC
group at the time of the first scan.
Volume Analysis of Longitudinal Brain MRI Scans
Comprehensive details of the longitudinal image processing are described by Aubert-
Broche et al.[7]. The pipeline is summarized here and is illustrated in Fig. 1.
For each time point of each subject, after the preprocessing steps (noise removal
[8], intensity inhomogeneity correction [9] and intensity normalization [10]), the T1w
images were linearly registered to a subject-specific linear template created using the
principles of average template construction described in Fonov et al.[10]. If a scan
had lesions, they were masked [11] to avoid matching healthy tissue to lesions. A
hierarchical nine-parameter linear registration based on an intensity cross-correlation
similarity measure was performed between the subject-specific linear template and
the ICBM152 template [12] to align the subject-specific template with the population
template.
Search WWH ::
Custom Search