Abstract
We present the results of the performance evaluation of an on-line interactive segmentation framework that addresses the problem of propagation of a single three dimensional (3-D) example to similar images. The aim is to minimise user intervention during interactive segmentation when an example of the desired outcome exists. The framework tackles the segmentation problem as a process consisting of two distinct tasks: the localisation of the region of interest and its boundary delineation. We have previously applied and evaluated this approach on the segmentation of the prostate with encouraging results. Here, we extend the evaluation study on additional datasets and application domains. For each stage of the framework, different candidates are evaluated against delineations provided by experts. The results suggest that a two stage strategy increases the framework's repeatability and that our approach can provide automated results close to the ground truth, allowing the user to intervene only if necessary at a later refinement stage or in case of provision of unsatisfactory segmentation outcomes.
Original language | English |
---|---|
Pages | 287-291 |
Number of pages | 5 |
Publication status | Published - 2011 |
Event | Medical Image Understanding and Analysis - Sheffield University Duration: 1 Jan 1824 → … |
Conference
Conference | Medical Image Understanding and Analysis |
---|---|
City | Sheffield University |
Period | 1/01/24 → … |
Keywords
- Interactive Segmentation Propagation
- Performance Evaluation
- Image Registration
- Graph-Cuts