Publication Type

Journal Article

Version

publishedVersion

Publication Date

7-2014

Abstract

In this paper, we propose a fast hierarchical framework of leukocyte localization and segmentation in rapidly-stained leukocyte images (RSLI) with complex backgrounds and varying illumination. The proposed framework contains two main steps. First, a nucleus saliency model based on average absolute difference is built, which locates each leukocyte precisely while effectively removes dyeing impurities and erythrocyte fragments. Secondly, two different schemes are presented for segmenting the nuclei and cytoplasm respectively. As for nuclei segmentation, to solve the overlap problem between leukocytes, we extract the nucleus lobes first and further group them. The lobes extraction is realized by the histogram-based contrast map and watershed segmentation, taking into account the saliency and similarity of nucleus color. Meanwhile, as for cytoplasm segmentation, to extract the blurry contour of the cytoplasm under instable illumination, we propose a cytoplasm enhancement based on tri-modal histogram specification, which specifically improves the contrast of cytoplasm while maintaining others. Then, the contour of cytoplasm is quickly obtained by extraction based on parameter-controlled adaptive attention window. Furthermore, the contour is corrected by concave points matching in order to solve the overlap between leukocytes and impurities. The experiments show the effectiveness of the proposed nucleus saliency model, which achieves average localization accuracy with F1-measure greater than 95%. In addition, the comparison of single leukocyte segmentation accuracy and running time has demonstrated that the proposed segmentation scheme outperforms the former approaches in RSLI.

Keywords

Average absolute difference, Cytoplasm enhancement, Leukocyte localization, Visual attention

Discipline

Software Engineering | Theory and Algorithms

Research Areas

Software and Cyber-Physical Systems

Publication

Micron

Volume

56

First Page

17

Last Page

28

ISSN

0968-4328

Identifier

10.1016/j.micron.2013.09.006

Publisher

Elsevier

Additional URL

https://doi.org/10.1016/j.micron.2013.09.006

Share

COinS