The alignment tool consists of five
main sections which apply from top to bottom to the order of operations
required to carryout a registration:
1.
Setup Approximate Rotations
2.
Current Transformation Estimate (Usually leave as is from setting above)
3.
Basic Auto-Alignment
4.
Transform and Save Floating Image to File
5.
Additional Alignment Tools
These
are described in more detail in the sections below. Steps 1 to 3 are generally required
for MRI-MRI, MRI-CT, MRI-PET and MRI-SPECT brain image registration
procedures.
1.
Setup Approximate Rotations
The first step is to set the overall
starting estimate for the transformation which will account for
differences in the acquisition types (axial/coronal/sagittal) etc. The
automated registration may be able to recover this automatically, but
its usually best to provide this manually. So.. Before starting the automated registration,
confirm that the images are displayed with the same anatomical slicing
angle (i.e. transaxial, coronal and sagittal planes roughly
correspond). If the images are not both transaxial, coronal or sagittal
then select appropriate starting estimate from the central panel of
buttons on the alignment tool.
If the images were both in DICOM or
NIFTI format then (from version 8.216B and later) you can select the
'From Data' button to caclulate the appropriate combination of 90
degree rotations to bring the images into approximate alignment. If
this does not work because the header data in the files did not supply
correct scan direction information, then you can manually select
a starting estimate: For example, to map a Coronal floating
image to a Transaxial Reference image volume:
select:
Which, for this data, gives:
If you are unsure of the original orientation of the data simply click
different buttons until the orthogonal slices look similar!
Alternatively, if the patient
orientation with respect to the data slices was read correctly from the
image formats (e.g. some dicom data) then use the 'From Data' button to
set this.
Note: If the image data format does not provide correct axis direction information , it may be the
case
that the images are reflected in one axis (for example top-bottom, or
more dangerously left- right). If this occurs please email me and I
will try and improve the file format reading!
2.
Current Estimate
This simply shows the
current rigid transformation parameters in terms of the three rotations
in degrees and three translations in mm. Each parameter may be manually
edited and after pressing enter/return in the parameter box the rview
display is updated with that transformation. Note: manual editing of
the starting estimate is not usually needed for whole brain data that
is not highly misaligned. It may be necessary when you have limited
anatomical extent in one or both of the image volumes (for example few
slices, or limited in-plane field of view).
Saving the
current transformation;
These controls also allow the
saving and loading of the current transformation parameters either n
the form of a 'DOF' file containing the translations and rotations or
in the form of a 4x4 transformation matrix. (all transformations are
with respect to mm co-ordinates in each image co-ordinate systems,
where
0,0,0 is at the centre of the imaged space). More details
of the form of these parameter files are given here.
3.
Basic Auto-Alignment
This set of controls
initiates the automated image alignment using maximisation of global normalised mutual information. Note: this
registration estimation simply refines the display parameters within
rview, and does not actually 'transform' the image (to do so see the
next section), since rview is capable of displaying affine
transformations on the fly.
To align a floating image
to the reference image, simply select the main button 'Align to Ref
Image'. This creates a background thread which resamples the images
into a pyramid (4D) and then sequentially aligns the pyramid levels in
order, to refine the alignment. During the alignment process rview
display updating and control proceeds as normal. When registration is
underway the status bar at the bottom of the main rview window
indicates 'Busy registering ...'
Registration time is very dependent on:
As the registration
proceeds, the rview display is updated every few seconds with the
current estimate, allowing the user to see if the registration is
proceeding correctly. When the registration is complete The 'Busy
Registering...' message in the lower status bar of the main window
disappears (if you have a speaker, there will also be a 'bleep' to
bring you back from coffee!). At this point you should use rview
display options to check the registration visually using either colour
overlay or iso-intensity contour displays of the floating image on the
reference image: refer to the step by step examples for the best way to
visually inspect registration.
Note: Because the registration thread is not run at the highest priority, registration speed may be influenced dramatically by other tasks running on the windows machine!). One example has been the presence of a virus process running on a users machine which captured all the CPU resources from the registration thread, preventing it from running... this took a while to track down!
Aligning multiple floating images
To align all floating
images (if you have many loaded) select the button 'Align All Floating
to Ref.' This will simply step through each
loaded floating image and run the automated alignment.
Speed/Accuracy Control
The accuracy and speed to the alignment step are basically determined
by the finest resolution of the image pyramid using in the
registration. This can be altered to three separate levels which
basically control the amount of memory available for the image pyramids
to be stored in. 'Std' is typically useful for MRI-PET/SPECT
registration where one or both of the image data sets are low
resolution (and or noisy). For more accurate registration, use the
'High' or 'Highest' options.
Transform
and Save to File
This simply allows the
floating image to be transformed and resampled to the Reference image
co-ordinates and saved to an image file, using different interpolation
techniques Selecting one of the options prompts for a file name (either
in analyze or gipl format). Note: For historical compatibility reasons
these are both in sun/little endian byte order... sorry!
Additional
Alignment Methods
These are provided for more
challenging image data:
'Refine Alignment' simply
re-runs the registration but only using the finest levels of the
pyramid. This allows local optima (for example set by manual
interaction) to be refined.
'Define Sub volume' This
allows the rigid alignment estimation to be estimated based only on a
sub volume of the reference image. This is useful where for example a
true rigid transformation is only really applicable to a subset of the
image data: for example where there is significant soft tissue
deformation between scans, around a region of interest, or where there
is geometric imaging distortion.
Clicking once on the button
activates the sub volume specification on the rview windows: Clicking
once in a window expands the sub volume (initially at the centre of the
reference image) to include the point clicked for example clicking once
with the right mouse button in the transaxial plane expanded the
sub volume in the axial plane:
The clicking in the sagittal plane expands this to a rectangular sub volume:
Any further automated
alignment steps will then use this sub volume for the finest level of
image alignment (note: coarse registration levels still use the whole
image for stability). The sub volume function may also be used with the
'refine' step to localize the registration.
To clear this and use the full image
again for all registrations: use 'Clear Sub volume'