An enhanced acute leukemia segmentation based on particle Swarm optimization

Abstract

One of the hemopoietic disorders in humans is acute leukemia. Cell growth in acute leukemia disease occurs rapidly and uncontrollably. Therefore, in order to maximize the efficacy of treatment, there is a need to detect the disease early. Recently, Computer-Aided Detection and Diagnosis (CAD) approaches have been developed to assist medical staff in interpreting medical images. A crucial CAD technique for the diagnosis and verification of diseases such as acute leukemia is image-segmentation. However, it is still challenging to segment acute leukemia cells from the background due to the inconsistency of intensity image for acute leukemia blood samples. The original acute leukemia image firstly utilizes the formula of saturation with reference to the colour spaces of the HSI. Subsequently, the S-component is obtained and fed into the PSO to perform the segmentation process. Besides that, in order to optimize the segmentation process and increase the detection accuracy, the k-means algorithm is proposed as the initial centroid for PSO, called hybrid k-means-PSO. The proposed methods are performed on 10 and 24 images of Acute Lymphocytic Leukemia (ALL) as well as Acute Myelogeneous Leukemia (AML) respectively, which have been captured using an Infinity2 camera mounted on Leica microscope. The k-means clustering are used as the reference standard for the performance evaluation. Simulation results indicate that both PSO and hybrid k-means-PSO methods have better accuracy compared to k-means with the highest accuracy obtained scoring up to 97.24% and 97.02% respectively. As a result, the proposed method can automatically segment acute leukemia cells from the background and is helpful for the classification stage.