Marching cubes: A high resolution 3D surface construction algorithm
SIGGRAPH '87 Proceedings of the 14th annual conference on Computer graphics and interactive techniques
Watersheds in Digital Spaces: An Efficient Algorithm Based on Immersion Simulations
IEEE Transactions on Pattern Analysis and Machine Intelligence
Semi-automatic generation of transfer functions for direct volume rendering
VVS '98 Proceedings of the 1998 IEEE symposium on Volume visualization
Hi-index | 0.00 |
The three-dimensional segmentation of regions of interest in medical images, be it a 2D slice by slice based approach or directly across the 3D dataset, has numerous applications for the medical professional. These applications may involve something as simple as visualisation up to more critical tasks such as volume estimation, tissue quantification and classification, the detection of abnormalities and more. In this paper we describe a method which aims to combine two of the more popular segmentation techniques: the watershed segmentation and the active contour segmentation. Watershed segmentation provides unique boundaries for a particular image or series of images but does not easily allow for the discrete nature of the image and the image noise. Active contours or snakes do possess this generalisation or smoothing property but are difficult to initialise and usually require to be close to the boundary of interest to converge. We present a hybrid approach by segmenting a region of interest (ROI) using a 3D marker-based watershed algorithm. The resulting ROI's boundaries are then converted into a contour, using a contour following algorithm which is explained during the course of the paper. Once the contours are determined, different parameter settings of internal/external forces allow the expert user to adapt the initial segmentation. The approach thus yields a fast initial segmentation from the watershed algorithm and allows fine-tuning using active contours. Results of the technique are illustrated on 3D colon, kidney and liver segmentations from MRI datasets.