Introduction to Active shape modelS
Overview
Active Shape Models (ASMs) is a method for image segmentation - very suitable for medical 2D and 3D images. An ASM is a statistical model that describes the shape variation of an object. The ASM algorithm is used to fit that model to the object in new unknown images.
Training
An ASM is constructed by using a training set of examples. The shape is represented by a set of points, so called andmarks. These landmarks are used to calculate the shape variation of the training set. By combining the shape information of the training set new shapes can be generated.
With a good training set the ASM can virtually take any possible shape of the object in question.
Fit the model
The ASM algorithm is used to fit the ASM to the object in new unknown images. The ASM iteratively deforms to fit the object but is not allowed to vary in a way that is not seen in the training set.
Fitting of the ASM
starts |
The ASM algorithm searches for the brain
surfaces and deforms the ASM for optimal match |
Finished! |
This unknown brain has severely reduced blood flow in large parts of the brain. The ASM algorithm finds the true brain shape and avoids uptake outside the of brain because the ASM only deforms as seen in the training set.
| Training examples |
Unknown image |
Some benefits
- Fast - a few seconds for 3D objects
- Deformable - but only deforms as seen in the training set
- Operates directly in 3D – not slice by slice
- No need for image warping - the model can be fitted directly to the image
- Can be used for 3D visualization
- Perfect for problem images where parts of the object are missing or with uptake outside of the object that could be mistaken to belong to the object
- The landmarks have corresponding anatomical positions – allows mapping to anatomical atlases