A robust algorithm for white blood cell nuclei segmentation

Segmentation of white blood cell nucleus is a crucial step in white blood cell counting and classification system based on peripheral blood smear images. It is also used in the automated diagnosis of blood cancer diseases. However, this step is a challenging task due to the variation of contrast and shape of the nucleus in peripheral blood smear images. This paper proposes an improved method for white blood cell nucleus extraction. The proposed method makes use of arithmetical operation guided by a control parameter, Fourier Transform algorithm for texture enhancement, mean shift technique for smoothing and boundary preservation, and k-means clustering algorithm with an adaptive K for nucleus extraction. The proposed segmentation algorithm was tested on 5 completely different image databases, and the results compared favorably with recent methods from good standing papers. An Average dice similarity coefficient of 97.35% was obtained for CellaVision database, 96.63% for normal leukocytes of ALL-IDB2 database, 93.48% for BloodSeg database, 93.14% for JTSC database, 88.63% for the healthy leukocytes of the ALL-IDB2 database and 86.02% for the LSCI database.

Automated summary

This paper proposes an improved method for white blood cell nucleus extraction, which achieves effective segmentation through a combination of multiple algorithms and techniques. The method performs well in tests on multiple databases and exhibits higher segmentation accuracy compared to other methods.