A novel coarse-to-fine adaptation segmentation approach for cellular image analysis

Cellular image content analysis is one of the most important aspects of the cellular research and often requires collecting a great amount of statistical information and phenomena. Automated segmentation of time-lapse images gradually becomes the key problem in cellular image analysis. To address fuzzy, irregular, and ruffling cell boundaries in time-lapse cellular images, this paper introduces a hierarchical coarse-to-fine approach which is composed of iteration-dependent adaptation procedures with high-level interpretation: initial segmentation, adaptive processing, and refined segmentation. The iteration-dependent adaptation lies in that the adaptive processing and the refined segmentation be deliberately designed without a fixed order and a uniform associated iteration number, to connect coarse segmentation and refined segmentation for locally progressive approximation. The initial segmentation could avoid over-segmentation from watershed transform and converge to some features using a priori information. Experimental results on cellular images with spurious branches, arbitrary gaps, low contrast boundaries and low signal-to-noise ratio, show that the proposed approach provides a close matching to the manual cognition and overcomes several common drawbacks from other existing methods applied on cell migration. The procedure configuration of the proposed approach has a certain potential to serve as a biomedical image content analysis tool.