Difference between revisions of "Projects:RegistrationLibrary:RegLib C02"

From NAMIC Wiki
Jump to: navigation, search
Line 56: Line 56:
 
*the defacing of the SPGR image introduces sharp edges that can be detrimental. We apply a multiresolution scheme at least. If this fails we mask that area or better still the brain. As a general rulle, if you have the mask available, use it.
 
*the defacing of the SPGR image introduces sharp edges that can be detrimental. We apply a multiresolution scheme at least. If this fails we mask that area or better still the brain. As a general rulle, if you have the mask available, use it.
 
*because of the contrast differences and the defacing we use '''Mutual Information''' as the cost function.
 
*because of the contrast differences and the defacing we use '''Mutual Information''' as the cost function.
 +
*a two-step approach may yield fast results: 1. alignment with default settings w/o mask, 2. alignment with higher sampling rate + mask + input transform from step 1
  
 
=== Registration Results===
 
=== Registration Results===

Revision as of 23:35, 16 December 2009

Home < Projects:RegistrationLibrary:RegLib C02

Back to ARRA main page
Back to Registration main page
Back to Registration Use-case Inventory

Slicer Registration Use Case Exampe: Intra-subject Brain MR FLAIR to MR T1

this is the fixed reference image. All images are aligned into this space this is a passive image to which the calculated transform is applied. It is a label-map in the same space as the moving FLAIR image lleft this is the moving image. The transform is calculated by matching this to the reference image this is a passive image to which the calculated transform is applied. It is a label-map in the same space as the moving FLAIR image LEGEND

lleft this indicates the reference image that is fixed and does not move. All other images are aligned into this space and resolution
lleft this indicates the moving image that determines the registration transform.
lleft this indicates images that serve as masks, i.e. they focus the active registration onto a specific area.
lleft this indicates images that passively move into the reference space, i.e. they have the transform applied but do not contribute to the calculation of the transform.

lleft T1 SPGR lleft mask lleft T2 FLAIR lleft segmentation
1mm isotropic
256 x 256 x 146
RAS
1mm isotropic
256 x 256 x 146
RAS
1.2mm isotropic
256 x 256 x 116
RAS
1.2mm isotropic
256 x 256 x 116
RAS

Objective / Background

This scenario occurs in many forms whenever we wish to align all the series from a single MRI exam/session into a common space. Alignment is necessary because the subject likely has moved in between series.

Keywords

MRI, brain, head, intra-subject, FLAIR, T1, defacing, masking, labelmap, segmentation

Input Data

  • Button red fixed white.jpgreference/fixed : T1 SPGR , 1x1x1 mm voxel size, sagittal, RAS orientation.
  • Button green moving white.jpg moving: T2 FLAIR 1.2x1.2x1.2 mm voxel size, sagittal, RAS orientation.
  • Button purple mask white.jpgmask: skull stripping labelmap obtained from SPGR, serves as mask
  • Button blue tag white.jpgtag: segmentation labelmap obtained from FLAIR.
  • Content preview: Have a quick look before downloading: Does your data look like this? SPGR Lighbox , FLAIR Lighbox
  • download dataset to load into slicer (~17 MB zip archive)

Registration Challenges

  • the amount of misalignment to be small. Subject did not leave the scanner in between the two acquisitions, but we have some head movement.
  • we know the underlying structure/anatomy did not change, but the two distinct acquisition types may contain different amounts of distortion
  • the T1 high-resolution had a "defacing" applied, i.e. part of the image containing facial features was removed to ensure anonymity. The FLAIR is lower resolution and contrast and did not need this. The sharp edges and missing information in part of the image may cause problems.
  • we have a skull stripping label map of the fixed image (T1) that we can use to mask out the non-brain part of the image and prevent it from actively participating in the registration.
  • we have one or more label-maps attached to the moving image that we also want to align.
  • the different series have different dimensions, voxel size and field of view. Hence the choice of which image to choose as the reference becomes important. The additional image data present in one image but not the other may distract the algorithm and require masking.
  • hi-resolution datasets may have defacing applied to one or both sets, and the defacing-masks may not be available
  • the different series have different contrast. The T1 contains good contrast between white (WM) and gray matter (GM) , and pathology appears as hypointense. The FLAIR on the other hand shows barely any WM/GM contrast and the pathology appears very dominantly as hyperintense.

Key Strategies

  • we choose the SPGR as the anatomical reference. Unless there are overriding reasons, always use the highest resolution image as your fixed/reference, to avoid loosing data through the registration.
  • the defacing of the SPGR image introduces sharp edges that can be detrimental. We apply a multiresolution scheme at least. If this fails we mask that area or better still the brain. As a general rulle, if you have the mask available, use it.
  • because of the contrast differences and the defacing we use Mutual Information as the cost function.
  • a two-step approach may yield fast results: 1. alignment with default settings w/o mask, 2. alignment with higher sampling rate + mask + input transform from step 1

Registration Results

  • the multi-resolution approach produces a very good registration even without the masking

Download