Optimised Feature Selection for Identification of Carcinogenic Leukocytes Using Weighted Aggregation Based Transposition PSO

In this paper, an automated leukocyte identification and classification methodology using microscopic blood smear images is proposed. After performing a colour based clustering for the identification of leukocytes, a large set of features are derived from the identified leukocytes. Thereafter a weighted aggregation-based transposition particle swarm optimisation technique (WATPSO) is proposed to select an optimised feature subset which is sufficient enough for classifying carcinogenic blast cells and normal cells efficiently. The proposed method is applied to a benchmark acute lymphoblastic leukaemia (ALL) data set and 97.30% test accuracy is achieved using only 6 optimal features. Experimental results demonstrate the usefulness of the proposed methodology in terms of high test accuracy and less number of informative selected features. The proposed method can be used as a decision support structure for haematologists for early diagnostic prediction of ALL.