A Novel Immersion Simulation Based Self-Organizing Transform with Application to Single-Cell Segmentation from Microscopy Images

Reliable cell segmentation plays an important role in biological imaging studies, though continues to be challenging due to the complex nature of many imaging scenes. The approach here uses a novel immersion simulation based self-organizing (ISSO) transform, an automated method for image segmentation. The method allows users to customize the immersion simulation process via user-defined or default self-organizing functions to incorporate prior information into segmentation. The transformed images usually have desirable features that can be further segmented using existing methods, with substantially improved segmentation results. The connections between the ISSO transform and existing watershed transforms is described. A Size Filter based on the ISSO transform is implemented and applied to various images and benchmark microscopy datasets from recent studies. With benchmark error rates well below those reported results in the literature, the comparison with other algorithms clearly demonstrates the benefits and flexibility of the ISSO method on various types of images.