Segmentation of CT and MRI datasets using 3D Slicer
From NCRR Biomedical Software Development, Engineering, and Dissemination Wiki
Contents |
Subproject Status
3D Slicer is an excellent tool for segmentation of medical imaging datasets.
Segmentation
Segmentation of images is the process of identifying and labeling regions of interest within an image. 3D Slicer is a powerful and robust tool for semi-automated and manual segmentation of images. It is built on the NAMIC toolkit, and arguably the open source standard for GUI based 3D image segmentation. It is currently optimized for segmentation of the MRI images of the brain, but can be used to segment any DICOM dataset.
Getting Started
The first step is to obtain the 3D Slicer program as detailed on the 3D Slicer site. 3D Slicer is available precompiled for many platforms or may built from source. You should familiarize yourself with the program by going through the excellent Tutorial link on the home page. There is also a wiki site with an example based tutorial here. The users email list is also worth subscribing to. Because 3D Slicer is an evolving tool somebody might have built or be building a tool you want.
Modifying Data
Once you are comfortable importing DICOM datasets as outlined in Tutorial turn your attention to the User's Guide linked on the same page. Within the User's Guide familiarize yourself with the Modifying Data section. The details, the tools, and the steps used to segment a DICOM dataset are described. Pay particular attention to the thresholding, save island, remove island, change island, and most importantly the all important UNDO button. The draw function will also be useful. Play with these until you are familiar with how they work.
Saving Data
At the end of the Modifying Data section it details the numerous ways to save data within 3D Slicer. Within this page and within 3D slicer a label map is defined as a voxelized volume. This can be saved in several formats but the most useful for our purposes is the NRRD formatcreated by Dr. Kindlmann. The NRRD format has an associated set of tools called unuwhich allow the manipulation and combination of individual NRRDs. The option to save as a NRRD file is a little hard to find. Within the Slicer window go to Volumes-->More and look down to the bottom where it sayes NRRD. Be sure to specify the appropriate label map for the 3D Slicer to save as a NRRD in the top Active Volume selector.
Another useful tool is to utilize the Modelmaker module to take your label map and convert it into a model which shows the surface of the label map in 3D and can be saved as a .vtk file. This is useful for examining your label maps as a surface within the 3D Slicer program.
These two formats should be understood as separate and useful for different purposes. The label maps saved in the NRRD format will be the building blocks which are combined and hierarchically "stacked" on one another to form the final model for import into SCIRun. The models saved as .vtk files are useful for visualization of the shell of the label map within the Slicer program.
Automated Segmentation
We are in the process of working to utilize segmentation algorithms developed for the brain at SPL for the chest, heart, and vasculature. Dr. Westin's group at SPL has substantial experience in segmentation of images and we are beginning to explore various techniques for automated segmentation.
Subproject Improvements/Requirements
- Automated segmentation techniques/algorithms
--MJ 09:42, 25 Jan 2006 (MST)
