are performed under Duke IACUCC protocol #A022 00 01 1. This is
a conventional perfusion fixation with inflow to left ventricle
and outflow from right atrium. The syringe pump is set at 9 ml/min.
The pressure at this setting and for the size of the tubing and
stub adapter, is about 75 mmHg. When setting up the pump, make
sure there are no bubbles in the line.
All scans are performed at 9.4 T using a 1 cm solenoid rf coil.
All scans are acquired in a single setting so the data sets are
All scans are acquired with a fixed field of view of 11x11x22
mm with a matrix of 256x256x512 yielding voxels of 0.043x0.043x0.043
mm i.e. 8 x 10-5 mm3.
We are acquiring three registered data sets using standard 3DFT
spin warp encoding. The TR and NEX have been adjusted so that
each set requires 14.56 hrs:
Data set A: T1 weighted @ TR=100
ms, TE=5.5 ms, Bandwidth ±31.25 kHz, NEX=8
Data set B: T2 weighted @ TR=200
ms, TE- 20 ms, Bandwidth ±15.62 kHz,NEX=4
Data set C: Spin density @ T2
weighted @ TR=200 ms, TE- 5.5 ms, Bandwidth ±15.62 kHz,NEX=4
When the Duke MRM is finished, the specimen is removed from the
fomblin, rinsed in 10% buffered formalin and placed in a sealed
container with formalin. The container is placed on chipped (wet)
ice and shipped overnight to Cal Tech where the diffusion tensor
imaging takes place. When the diffusion tensor imaging is completed,
the specimen is forwarded to UCLA.
Diffusion Tensor Imaging
Magnetic resonance imaging is done at 37° C in an 11.7 Tesla,
vertical bore (89 mm) Bruker AMX500 microimaging system (Bruker
Instruments). We use an Acustar shielded gradient set (max 100
gauss/cm gradient strength) with home-built RF probes and low-noise
preamplifier. Images are recorded using 3D multispin echo protocols
(one to eight echoes) with a data matrix of 256 x 128 x 128 points.
Typical spatial resolution is approximately 60 µm3 per voxel.
The images are padded with zeros to double the number of time
domain points in each dimension, the Fourier transformed to yield
a matrix of 512 x 256 x 256. This procedure is commonly called
"zero-filling" and is a well known interpolation method
Blockface and Histology
After post-fixation the brains are dipped in a mixture of india
ink (Pelikan) and 5% gelatin (Sigma) to simplify segmentation
of tissue from background later. The brains are then embedded
in OCT compound (Sakura) at 4° C and snap-frozen at -70°
C in a 2-methylbutane/dry ice bath.
Once the blockface and histological samples are prepared, the
brains are attached to a chuck with OCT compound (Sakura), and
sections are cut serially in 50µm thick coronal sections
on a modified CM3050S cryostat (Leica). A DMCIe digital camera
(Polaroid) captures images of the blockface prior to each section
at a resolution of 1600 x 1200 (approximately 6.7µm/pixel)
in 24-bit color. Sections 200µm apart are Nissl-stained
(Thionin) and alternating sections 200µm apart are myelin-stained
using a modified myelin impregnation stain. Stained preparations
are digitized using a 0.5X objective on an AX70 microscope (Olympus)
with a DMCIe digital camera (Polaroid) at a resolution of 1600
x 1200 (approximately 6.1µm/pixel) in 24-bit color. The
images are acquired using a macro imaging system that provides
undistorted high-resolution images with even illumination across
the entire field of view.
The method we use to extract mouse brain tissue from the embedding
medium used during cryosectioning and the slide background for
histologically stained sections takes advantage of the rich textural
information of the images, for example, the texture of the frozen
OCT compound is quite different from that of the brain tissue.
A previously trained neural net classifies a region within the
image as either tissue or background based on the textural descriptors
and output a segmented image. Combined with color information,
it provides sufficient discriminating power to perform the segmentation.
The two dimensional digital images of the stained sections are
brought roughly into register with their corresponding blockface
images acquired during sectioning using automated software tools
produced at LONI. The preregistration program produces an initialization
file for Automated
Image Registration (AIR). Once the registered images are reconstructed
into a three-dimensional volume, the resulting volume is brought
into register with an inherently three-dimensional uMRI in a defined
and common coordinate system. All image processing is done on
a 32-processor Onyx 200 supercomputer (SGI).