Publication details
Topology Preserving Segmentation Fusion for Cells with Complex Shapes
| Authors | |
|---|---|
| Year of publication | 2021 |
| Type | Article in Proceedings |
| Conference | The IEEE International Symposium on Biomedical Imaging |
| MU Faculty or unit | |
| Citation | |
| Doi | http://dx.doi.org/10.1109/ISBI48211.2021.9433867 |
| Keywords | Segmentation fusion; Reference annotation; Cell annotation; Shape |
| Description | We present an algorithm to fuse simply connected segmentation masks of complex and variable shapes that often appear in cell imaging. The algorithm is designed to preserve topology of the input masks and to faithfully represent their protrusions. It works in three main phases: (1) the detection of geodesic ends that correspond to the protrusions, (2) optimal matching of the geodesic ends, and (3) contour averaging of corresponding boundary segments. We show that our algorithm overcomes commonly used pixel-wise fusion algorithms (namely majority voting, SIMPLE, STAPLE, and topology-preserving STAPLE), as well as recently published geometric median shapes in terms of the visual quality of results as well as better representation of protrusions. We demonstrate the performance of our method based on synthetic images as well as real images from the cell segmentation benchmark datasets. |
| Related projects: |