Robust Graph Pruning for Efficient Segmentation and Cluster Splitting of Cell Nuclei Using Deformable Shape Models

We introduce a new efficient and robust method for cell nuclei segmentation and cluster splitting, which is based on implicitly parameterized deformable shape models and convex optimization. We propose a novel graph pruning algorithm for cluster splitting, which leads to graphs with fewer nodes compared to previous work and improves the computational efficiency. In addition, we prove that the approximation error introduced by graph pruning is bounded, meaning that the algorithm is intrinsically robust. We evaluated our method using fluorescence microscopy images from four public datasets and found that our method yields state-of-the-art results and strongly improved computational efficiency.