An Unsupervised and Supervised Combined Approach for White Blood Cells Segmentation

The white blood cell (WBC) segmentation and classification is a challenging task, due to the different shapes of the nucleus, cytoplasm and the number of lobes. The purpose of this paper is to provide a method for fast and accurate segmentation of leukocyte in smear images by a convolutional neural network (CNN) model and Gaussian Mixture Model (GMM) approach. The first step is the usage of white balance and selfdual multiscale morphological toggle (SMMT) to increase the contrast between the nucleus and cytoplasm. To segment, each WBC and corresponded nucleus and cytoplasm regions, a CNN model with 10 layers and GMM are used, respectively. In the postprocessing step, removing undesired objects by size, closing, and filling morphological operations are applied to each segment. The proposed method is validated on peripheral smear blood images in Cellavision dataset. This dataset contains 27 images which include different types of normal leukocytes. In order to evaluate the proposed method, the Dice coefficient, Jaccard and F1-score are used. The experimental results demonstrate the high accuracy for segmentation results of different types of WBC.