Introduction
Segmentation Software
There is a wide range of paid software that is capable of performing segmentation operations, but these freely available packages combine ease of use with a good range of functionality
3D Slicer:
•Available from https://www.slicer.org/
•A free software that is primarily designed for visualization of 3D medical image data •It also has some features for image analysis, including registration and segmentation •In our 3D Slicer worked example you can use this software to produce a segmentation of the lung |
Seg3D:
•Available from http://www.sci.utah.edu/cibc-software/seg3d.html
•This free software is a more advanced volume extraction program with both manual and automatic segmentation tools •Uses Layer based approach to allow easy manipulation of multiple segmentations from a set of image data •In our Seg3D worked example, you use this software to create a rib segmentation |
Segmentation Techniques
Segmentation as a field is always advancing in response to new imaging technology and novel applications. Even the simplest segmentation software will have a range of basic tools and advanced algorithms to extract structures. Here are some examples of the most commonly used techniques to get you started in developing your segmentation skills.
Basic:
Thresholding
•Pixels are partitioned depending on their intensity value. This effectively converts a grayscale image to a binary image.
•Pixels are partitioned depending on their intensity value. This effectively converts a grayscale image to a binary image.
Advanced:
Parametric models – Snakes
•The algorithm attempts to model the edges by minimising an energy term. This is minimised when the contour is on the object boundary and when the contour is as regular and as smooth as possible.
•It is useful for interpreting incomplete images and is robust to noise, but it can be slow to compute.
•The algorithm attempts to model the edges by minimising an energy term. This is minimised when the contour is on the object boundary and when the contour is as regular and as smooth as possible.
•It is useful for interpreting incomplete images and is robust to noise, but it can be slow to compute.
Expectation Maximisation (EM)
•The algorithm finds the maximum likelihood of label distribution in a probabilistic way.
•This framework is highly complex but can be a powerful tool for modelling the data accurately.
•The algorithm finds the maximum likelihood of label distribution in a probabilistic way.
•This framework is highly complex but can be a powerful tool for modelling the data accurately.