Morphological Filtering and Hierarchical Deformation for Partially Overlapping Cell Segmentation

Accurate cell segmentation is an important and long-standing challenge in biomedical image analysis due to large variations in shape and boundary ambiguity. In this paper, we present a segmentation framework for partially overlapping cervical cells. The proposed method starts by cellular clump estimation with morphological reconstruction. Subsequently, the nuclei inside the cellular clumps are located by H-maxima transformation and thresholding. The cytoplasm of each detected nucleus is finally delineated with hierarchical deformation based on landmarks and shape dictionaries. The proposed approach is tested on a cervical smear image dataset containing single and partially overlapping cells. The results demonstrate that our approach can achieve more accurate and stable cytoplasmic segmentation, better nuclear segmentation, and lower time complexity, compared to a state-of-the-art approach.