- 1 Slicer Registration Training Videos
- 1.1 Registration Masking: How do I quickly generate a mask for use in registration?
- 1.2 Registration Masking: How accurate should my mask be ?
- 1.3 Registration Validation How do I check / visualize a registration result ?
- 1.4 Fiducial Based Registration: How can I register two volumes with manually selected landmarks?
Slicer Registration Training Videos
Below you will find brief videos/screencasts showing step-by-step approaches to the most common tasks. Most have no audio.
Registration Masking: How do I quickly generate a mask for use in registration?
There are several segmentation and editor tools that you can use to quickly generate a map to focus a registration onto the important content. Six approaches are shown below, but not all work equally well for all images. Example image is a T2-weighted MRI with a tumor. We seek to mask the brain parenchyma, i.e. perform basic skull stripping. We show these to give an introduction and overview over the tools available and the general approach in using them. Many variations and combinations exist. We therefore recommend you try a few on your data to get a feel of what works best for your needs. Also remember that masks for registration differ from other segmentation in that they do not need to be highly accurate. Also masks should include the dominant edges, thus it is recommended to dilate the final segmentation by a few pixels (see separate topic below).
- Subsample input image (1 min): One effective way to reduce manual editing time is to create a subsampled version of your data before you start, perform the segmentation/editing on the subsampled mask and then upsample again in the end. This works best for masks of structures that are large and have relatively smooth surfaces.
- Otsu automated thresholding (2 min): This fast and simple method works well if you seek to separate an object from the background. This will require some post-editing to cleanup. Not recommended for small structures.
- Robust Statistics Segmenter (3 min): This is a very effective segmenter that requires an initial "rough" segmentation as a starting point. You control the result via parameters for the smoothness, size and expected intensity variation of your object. Quite efficient in combination with level sets, as shown in the video.
- Grow-Cut Segmenter (4 min): This is a very powerful algorithm with sparse controls. Fast and effective. It also requires an initial "rough" segmentation as a starting point. You provide sample regions of at least 2 structures and let the module extrapolate. You can then reiterate by applying manual corrections and rerun. Very effective in combination with level sets, as shown in the video.
- Threshold Paintbrush (3 min): This editor tools combines the standard paintbrush with a threshold effect. You paint your ROI but only pixels within a specified threshold are selected. This is very useful to trace complex edges, which you can pick up effectively with a large brush in a few strokes.
- Global Threshold (2 min): This editor tool is the standard global threshold applied to the entire image. This simple approach can be very efficient if your structure of interest has good contrast and separates well spatially. Morphological filtering can be applied to clean up the result.
- Region Growing (3 min): This tool requires you to select a few landmarks/fiducial points inside your object and will then try to grow outward looking for edges. Works best if your object has strong edges and is not connected to neighboring structures. It's fast enough for you to quickly try multiple parameter settings, but results will vary. Often one of the above tools will be more effective.
- Upsample label map (2 min): This is the post-editing counterpart to the subsampling above. Once you complete the manual editing, resample the labelmap back to its original resolution. Includes smoothing techniques.
- All of the above (21 min): Joint video showing all of the above steps in succession.
- New: SwissSkullStripper : this is a new extension module that can generate quick and fairly accurate skull stripping masks. Movie shows installation and use
- New:Magic Wand Tool : this is a new editor effect that can be very useful in generating quick masks for very irregular structures in a highly interactive way. Movie shows example use of generating a mask on a mouse brain MRI.
Registration Masking: How accurate should my mask be ?
A highly accurate segmentation can be counter-productive when used as a mask in registration. If it follows the main edges of the structure of interest (e.g. brain surface) exactly, then that edge is excluded from actively driving the registration. The result will be an incorrect excessive amount of expansion or shrinkage near the edge. We therefore recommend to dilate your mask a few pixels beyond the edges. The video below shows the difference:
Registration Validation How do I check / visualize a registration result ?
Visualizing complex datasets is one of 3DSlicer's great strengths. There are several tools/methods you can use to assess the result of a registration. The most common ones are shown below:
- Install & use Deformation Field Visualizer Extension (3 min, 20MB): For visualizing deformation (BSpline) fields the SlicerRT extension provides an extensive interactive tool that renders the deformation field as both slice projections and 3D models. Movie shows how to install & use. You will need a recent Slicer build,i.e. June 2013 or after.
Fiducial Based Registration: How can I register two volumes with manually selected landmarks?
There are two fiducial registration tools available within Slicer. The default is within the Registration menu, and an alternative version is found in the SlicerIGT extension as "Fiducial Registration Wizard". A detailed case example using both tools is available in the Case 48 in the Registration Case Library. A brief screencast on selecting landmarks and running either tools are also here:
- fiducial-based registration
- alternative tool: using SlicerIGT's fiducial-based registration (shows extension download and module execution only, for the landmark selection see movie above)