Getting started#
What is model-driven registration?#
Model-driven registration is a method to remove motion from a series of 2D or 3D images. It applies specifically to situations where a model exists that can describe the changes in signal intensity through the series.
Background#
Many applications in medical imaging involve time series of 2D images or 3D volumes that are corrupted by subject motion. Examples are T1- or T2- mapping in MRI, diffusion-weighted MRI, or dynamic contrast-enhanced imaging in MRI or CT.
Motion correction of such data is challenging because the signal changes caused by the motion are superposed on the often drastic changes in intrinsic image contrast. However in most cases these intrinsic changes in image contrast can be described by a known signal model. Indeed many of these applications critically depend on the availability of a model to derive parametric maps from the signal data.
Model-driven image registration leverages the existence of a signal model to remove the confounding effects of changes in image contrast on the results of the motion correction.
Any model-driven registration method therefore requires two ingredients:
a signal model that describes the changes in signal in the absence of motion.
a motion model that describes the changes in signal caused by motion alone.
mdreg has a library of built-in signal models for common scenarios,
also includes a simple mechanism for integrating custom-build signal models.
If a model is not specified, mdreg will assume there are no intrinsic
changes in intensity and will use a constant signal model.
For motion modelling, mdreg offers a unified interface to coregistration
methods from different packages - including itk-elastix, scikit-image
and dipy. If a coregistration method is not specified, mdreg will
use bspline coregistration fom the package elastix
How to use mdreg?#
The getting started section in tutorials illustrates
different types of usage, and is a good place to start if you have not
used mdreg before.
[.. more coming soon..]