Shape-Aware Cell Segmentation with Global Cues
| Authors | |
|---|---|
| Year of publication | 2026 |
| Type | Peer-reviewed scientific article |
| Magazine / Source | IEEE SIGNAL PROCESSING LETTERS |
| MU Faculty or unit | |
| Citation | |
| web | https://doi.org/10.1109/LSP.2026.3721049 |
| Doi | https://doi.org/10.1109/LSP.2026.3721049 |
| Keywords | Cell segmentation; deep learning; convolutional neural network; implicit active contours; shape-related cues |
| Description | While convolutional neural networks are still widely used in bioimage segmentation tasks, they suffer from an inherent issue: lack of shape awareness and long-range context due to processing image data exclusively using local operations. In response, we present an architecture that joins CNNs with implicit active contours, deliberately designed to perform segmentation with global and shape-related cues at sub-pixel accuracy. Quantitative validation using three cell imaging modalities and three complementary measures shows that our method is superior to its CNN- and transformer-based competitors as well as a pretrained generalist Cellpose-SAM model: our method decreases boundary localization errors by as much as 82% while achieving reliable overall performance. |
| Related projects: |