Automated identification of plasma cell in bone marrow images

Abstract

The classification and the count of different White Blood Cells (WBC) in microscopy images allow the assessment of a wide range of important hematology pathologies such Myeloma. This article aims particularly myeloma disease; this pathology is manifested by a proliferation of a type of cell called plasma cells. This paper presents a robust and accurate novel method for processing WBC(leukocyte) using a combination of ideas. The segmentation of cells is achieved in two phases, the first is to extract the nucleus from the green channel of the image using the otsu's thresholding method, and the second operates through a region growing on the nucleus in order to delimitate a cytoplasm. After that, a cell's characterization step is performed where desired features can be extracted in order to classify types of leukocytes and thus allow the counting of plasma cells. The proposed process have been successfully tested to a large number of images, showing promising results in terms of segmentation and classification of plasma cells, and also applied to varying cell appearance and image quality, encouraging future works.

Multi-feature-based approach for white blood cells segmentation and classification in peripheral blood and bone marrow images

“…As shown in Table 4, an average accuracy of 96.87% and 92.50% was obtained for nucleus and cytoplasm segmentation, respectively. Therefore, the proposed method achieved better results than Benazzouz et al (2013Benazzouz et al ( , 2015Benazzouz et al ( , 2016 especially in cytoplasm regions. Indeed, our segmentation extracts the cytoplasm regions precisely even when the shape boundaries are irregular.…”

“…Indeed, our segmentation extracts the cytoplasm regions precisely even when the shape boundaries are irregular. We can note that the circularity criterion that prevents the deformation of the region growing in Benazzouz et al (2015) and the misclassification between RBCs and some cytoplasm regions in Benazzouz et al (2013Benazzouz et al ( , 2016 affect the segmentation accuracy.…”

“…Even the diagnostic of experts is based essentially on this type of cells. For a quantitative evaluation, we compare the performance of our experiments to the methods mentioned in the related work on the same images dataset (Benazzouz et al, 2013(Benazzouz et al, , 2015(Benazzouz et al, , 2016. As shown in Table 4, an average accuracy of 96.87% and 92.50% was obtained for nucleus and cytoplasm segmentation, respectively.…”

“…A first improvement with respect to the method described earlier by Benazzouz et al (2015), where the authors separate WBCs into two classes (normal and dystrophic plasma cells vs. other types) in order to diagnose myeloma pathology, lies in the number of WBC classes that the proposed method can distinguish. The obtained plasma cells accuracy in their work is 75.25% due to the misclassification of most dystrophic plasma cells into other cells type.…”

“…Firstly, nucleus extraction is performed by Otsu thresholding from green channel, then a region growing with circularity criterion delimitates the cytoplasms. Features extraction and cells classification is presented in Benazzouz et al (2015). Segmentation scheme using pixel classification based on the fusion of information and evidential algorithm to segment blood cell images is reported in Benazzouz et al (2013Benazzouz et al ( , 2016 and Baghli et al (2014).…”