An efficient algorithm for detection of white blood cell nuclei using adaptive three stage PCA-based fusion
Mimosette Makem, Alain Tiedeu
Abstract
Morphological analysis of white blood cells under a microscope is a crucial laboratory procedure for the diagnosis of several diseases including leukemia. This process is however, tedious and subjective, hence the need for innovative, efficient, and automatic analytical techniques. From shortcomings of algorithms in the literature such as inability to segment images with poor brightness and low contrast or deterioration of morphological structure of nucleus during segmentation, this paper sets forth an automatic method for segmentation of white blood cell nuclei in peripheral blood smear images in an attempt to cover these defects. In the proposed method, an adaptive fusion based on principal components analysis is used to fuse color components derived from HSV and CMYK color spaces. The proposed method enables the extraction of the nuclei without the application of morphological operations and presents the advantage of accentuated contrast thanks to PCA-based fusion and independence to color and brightness variation in the peripheral images. The developed algorithm was tested on three completely different public image datasets. The comparison of the results of the proposed algorithm with several other existing methods on the same datasets demonstrated its greater efficacy. A dice similarity coefficient of 97.06% was obtained for CellaVision, 94.75% for BloodSeg base, and 90.79% for JTSC databases.