Intelligent leukaemia diagnosis with bare-bones PSO based feature optimization

• Published in: Applied soft computing 2017-07, Vol.56, p.405-419
• Main Authors: Srisukkham, Worawut, Zhang, Li, Neoh, Siew Chin, Todryk, Stephen, Lim, Chee Peng
• Format: Article
• Language: English
• Subjects: Acute lymphoblastic leukaemia classification; Bare-bones particle swarm optimization; Feature selection
• ISSN: 1568-4946; 1872-9681
• DOI: 10.1016/j.asoc.2017.03.024

[Display omitted] •We propose BBPSO-based feature optimization for leukaemia diagnosis.•Two evolutionary BBPSO algorithms are proposed.•The first algorithm incorporates accelerated food chasing and flee mechanisms.•The second algorithm exhibits these two new operations in subswarm-based search.•They outperform other PSO variants and related research for leukaemia diagnosis. In this research, we propose an intelligent decision support system for acute lymphoblastic leukaemia (ALL) diagnosis using microscopic images. Two Bare-bones Particle Swarm Optimization (BBPSO) algorithms are proposed to identify the most significant discriminative characteristics of healthy and blast cells to enable efficient ALL classification. The first BBPSO variant incorporates accelerated chaotic search mechanisms of food chasing and enemy avoidance to diversify the search and mitigate the premature convergence of the original BBPSO algorithm. The second BBPSO variant exhibits both of the abovementioned new search mechanisms in a subswarm-based search. Evaluated with the ALL-IDB2 database, both proposed algorithms achieve superior geometric mean performances of 94.94% and 96.25%, respectively, and outperform other metaheuristic search and related methods significantly for ALL classification.