Wavelet-initialized 3D level-set cell segmentation with local background support

Segmentation is a core element of computational microscopy since it enables the detailed outlining of individual cells and cell-level measurements of various reactions to drug compounds. Many biological experiments also require confocal imaging to enable imaging of structures in 3D, and such experiments result in large 3D volumes stored as image stacks that can require significant computer memory and computation time to process. We propose several important extensions to a fast level-set algorithm that evolves multiple level-sets quickly and in a memory efficient manner for fast and accurate 3D segmentation. First, we integrate a wavelet segmentation step with the level-set algorithm to take advantage of the complementary advantages of the two methods. Second, we introduce a method whereby cells in backgrounds with spatially-varying intensity can be segmented without requiring a flat-field correction step. Third, we introduce a neighbor detection algorithm for evolving multiple level-sets without merging that is 85X faster than the topological number method. We demonstrate the significant improvements in accuracy and speed of these algorithms on multiple 3D datasets.