Projects:RegistrationLibrary:RegLib C03

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

Slicer Registration Library Exampe #3: Diffusion Weighted Image Volume: align with structural reference MRI

				LEGEND this indicates the reference image that is fixed and does not move. All other images are aligned into this space and resolution this indicates the moving image that determines the registration transform. 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.
T2	mask		DTI Baseline	DTI volume
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 is a typical example of DTI processing. Goal is to align the DTI image with a structural scan that provides accuracte anatomical reference. The DTI contains acquisition-related distortion and insufficient contrast to discern anatomical detail.

Keywords

MRI, brain, head, intra-subject, DTI, DWI

Input Data

• reference/fixed : T2w FSE,
• moving: Baseline image of acquired DTI volume, corresponds to T2w MRI , 0.9375 x 0.9375 x 1.4 mm voxel size, oblique
• 20px moving: Tensor data of DTI volume, oblique, same orientation as Baseline image. The result Xform will be applied to this volume.

Registration Results

Download

• download entire package (Data,Presets,Tutorial, Solution, zip file 33.7 MB)
  • download registration parameter presets file (MRML file, import as scene)
  • download image dataset only (NRRD, 10.7 MB, filename: RegLib_C01_Data_TumorGrowth.zip)
  • download image dataset only (NRRD, 10.7 MB, filename: RegLib_Case_01_NRRD_TumorGrowth.zip)
  • download image dataset in NIFTI format (NIFTI / nii, 10.7 MB, filename: RegLib_C01_DataNIFTI_TumorGrowth.zip)
  • result transform file (ITK .tfm file, load into slicer and apply to the target volume)
  • Tutorials (step-by -step walk through):
    • download/play video tutorial (quicktime, 15.9 MB, filename: RegLib_C01_VideoTutorial_TumorGrowth.mov)
    • download power point tutorial (zip file, 2.8 MB, filename: RegLib_C01_PPTTutorial_TumorGrowth.ppt.zip)
    • download step-by step text instructions (rtf text file)
• Multiresolution testresult package (Data,Xform, Solution, zip file 16.5 MB)
• Download package from MIDAS server (Data,Xform)

Discussion: Registration Challenges

• accuracy is the critical criterion here. We need the registration error (residual misalignment) to be smaller than the change we want to measure/detect. Agreement on what constitutes good alignment can therefore vary greatly.
• the two images have strong differences in coil inhomogeneity. This affects less the registration quality but hampers evaluation. Most of the difference does not become apparent until after registration in direct juxtaposition. Bias field correction beforehand is recommended.
• we have slightly different voxel sizes
• if the pathology change is substantial it might affect the registration.

Discussion: Key Strategies

• the two images have identical contrast, hence we consider "sharper" cost functions, such as NormCorr or MeanSqrd
• general practice is to register the follow-up to the baseline. However here the follow-up has slightly higher resolution. From an image quality/data perspective it would be better to use the highest resolution image as your fixed/reference. But here follow the most common convention, i.e. fixed image is the baseline.
• because we seek to assess/quantify regional size change, we must use a rigid (6DOF) scheme, i.e. we must exclude scaling.
• if the pathology change is soo large that it might affect the registration, we should mask it out. The simplest way to do this is to build a box ROI from the ROItool and feed it as input to the registration. Remember that masking does not mean that masked areas aren't matched, they just do not contribute to the cost function driving the registration, but move along passively. Next more involved level would be to outline the tumor. If a segmentation is available, you can use that.
• these two images are not too far apart initially, so we reduce the default of expected translational misalignment
• because accuracy is more important than speed here, we increase the sampling rate from the default 2% to 15%.
• we also expect minimal differences in scale & distortion: so we can either set the expected values to 0 or run a rigid registration
• we test the result in areas with good anatomical detail and contrast, far away from the pathology. With rigid body motion a local measure of registration accuracy is representative and can give us a valid limit of detectable change.