Projects:SlicerFAQ:RegistrationRerun

• back to Slicer FAQ

Automated Registration Result is Insufficient

• Problem: automated registration provides an alignment that is insufficient, possibly worse than the initial position
• Explanation: The automated registration algorithms (except for fiducial and surface registration) in Slicer operate on image intensity and try to move images so that similar image content is aligned. This is influenced by many factors such as image contrast, resolution, voxel anisotropy, artifacts such as motion or intensity inhomogeneity, pathology etc, the initial misalignment and the parameters selected for the registration.
• Fix: Your first try however should be to obtain a better automated registration by changing some of to re-run the automated registration, while changing either initial position, initialization method, parameters or the method/module used. Most helpful to determine a good secondary approach is to know why the first one was likely to fail. Below a list of possible reasons and the remedies:
  • too much initial misalignment: particularly rotation can be difficult for automated registration to capture. If the two images have strong rotational misalignment, consider A) one of the initialization options (e.g. BRAINSfit or Expert Automated), B) a manual initial alignment using the Transforms module and then use this as initialization input
  • insufficient detail: consider increasing the number of sample points used for the registration, depending on time/speed constraints, increase to 5-10% of image size.
  • insufficient contrast: consider adjusting the Histogram Bins (where avail.) to tune the algorithm to weigh small intensity variations more or less heavily
  • strong anisotropy: if one or both of the images have strong voxel anisotropy of ratios 5 or more, rotational alignment may become increasingly difficult for an automated method. Consider increasing the sample points and reducing the Histogram Bins. In extreme cases you may need to switch to a manual or fiducial-based approach
  • distracting image content: pathology, strong edges, clipped FOV with image content at the border of the image can easily dominate the cost function driving the registration algorithm. Masking is a powerful remedy for this problem: create a mask (binary labelmap/segmentation) that excludes the distracting parts and includes only those areas of the image where matching content exists. This requires one of the modules that supports masking input, such as: BRAINSFit, ExpertAutomated, Multi Resolution. Next best thing to use with modules that do not support masking is to mask the image manually and create a temporary masked image where the excluded content is set to 0 intensity; the Mask Volume module performs this task
  • too many/too few DOF: the degrees of freedom (DOF) determine how much motion is allowed for the image to be registered. Too few DOF results in suboptimal alignment; too many DOF can result in overfitting or the algorithm getting stuck in local extrema, or a bad fit with some local detail matched but the rest misaligned. Consider a stepwise approach where the DOF are gradually increased. BRAINSfit and Expert Automated provide such pipelines; or you can nest the transforms manually. A multi-resolution approach can also greatly benefit difficult registration challenges: this scheme runs multiple registrations at increasing amounts of image detail. The Robust Multiresolution module performs this task.
  • inappropriate algorithm: there are many different registration methods available in Slicer. Have a look at the Registration Method Overview and consider one of the alternatives. Also review the sortable table in the Registration Case Library to see which methods were successfully used on cases matching your own.
  • you can adjust/correct an obtained registration manually, within limits, as outlined here.