Plasma cell identification based on evidential segmentation and supervised learning

Abstract

Myeloma disease is among the most common type of cancer, it is characterised by proliferation of plasma cells, kind of white blood cell (WBC). Early diagnosis of the disease can improve the patient's survival rate. The manual diagnosis involves clinicians to visually examine microscopic bone marrow images for any signs of cells proliferation. This step is often laborious and can be highly subjective due to clinician's expertise. Automatic system based on WBC identification and counting provides more accurate result than manual method. This system is mainly based on three major steps: cell's segmentation, cell's characterisation and cell's classification. In the proposed system, microscopic images of bone marrow blood are segmented by the combination of watershed transform and the evidence theory, the segmented cells are characterised with shape and colour texture features, and then classified into plasma cells or not plasma cells with three supervised classifiers; support vector machines, K-nearest neighbour and decision tree. Experimental results show that the recognition of plasma cells with the K-nearest neighbour achieved 97% of correct rate with 100% of specificity.