Wrapping up: Week 7

In this week we completed our toolchain with an algorithm for fitting a shape model to an image. We have seen the classic algorithm called Active Shape Model fitting.

Active Shape Model fitting turned out to be just a variant of the ICP strategy we saw last week, with the difference that we do not select the closest point as the corresponding point in each iteration, but the point in the image that best fits a learned intensity model.

Being able to fit a model to an image is immensely useful. A successful fit of a model to an image implies, for example, a solution to the image segmentation problem, which is one of the most important problems in medical image analysis. But given a fit of the model we have much more information than what is required for segmentation: we know for every point of the model to which point of the image it corresponds. Hence, we can transfer any information labelled on the reference surface, such as landmarks, anatomical regions, and so on directly onto the target image. We have analysed the image in terms of our model.

To conclude this section we would like to reiterate a point we already made last week. If we have successfully fitted a model, be it to a surface or to an image, we get a solution to many important problems almost for free. This implies, of course, that model fitting itself is a really difficult problem. While the algorithms presented here work surprisingly well in many applications, the problem of model fitting is far from being solved. Indeed there is a lot of research going on and new solution are being proposed continuously.

In the next and final week of this course we will give you some examples of what can be done with a successfully fitted model, and point you in directions where you can continue your learning journey.

Share this article:

This article is from the free online course:

Statistical Shape Modelling: Computing the Human Anatomy

University of Basel