Smart Seed Selection Using Inverse Ridge Response for Region Growing Segmentation

Automated cell segmentation using a fast and accurate algorithm could be a convenient tool for the biologist to further analyze the data, enabling corresponding statistical inference. Though there are lots of segmentation techniques available, accurate segmentation is difficult and typically requires problem-specific tailoring of algorithms. In this paper, we propose a semi-supervised segmentation algorithm having four main input parameters that could be easily estimated using prior biological knowledge of the cell type and imaging modality. The method utilizes seeds for the post-region growing-based segmentation. The proposed algorithm consists of ridge-based filtering and local smoothing in sequence to help find the regional maxima ensuring one for each cell. These regional maxima are considered as the initialization of region-growing segmentation. The suggested method is faster, lightweight, and more vigorous compared to state-of-the-art algorithms, hence called 3S (Smart Seed Selection). Deep learning-based methods may outperform the suggested method in accuracy in some cases. The recommended approach is general enough to be used with lighter calculations across a wide variety of cell types and imaging modalities.