Acute lymphoblastic leukemia segmentation using local pixel information

Abstract

The severity of acute lymphoblastic leukemia depends on the percentages of blast cells (abnormal white blood cells) in bone marrow or peripheral blood. The manual microscopic examination of bone marrow is less accurate, time-consuming, and susceptible to errors, thus making it difficult for lab workers to accurately recognize the characteristics of blast cells. Researchers have adopted different computational methods to identify the nature of blast cells; however, these methods are incapable of accurately segmenting leukocyte cells due to some major disadvantages, such as lack of contrast between objects and background, sensitivity to gray-scale, sensitivity to noise in images, and large computational size. Therefore, it is indispensable to develop a new and improved technique for leukocyte cell segmentation. In the present research, an automatic leukocyte cell segmentation process was introduced that is based on machine learning approach and image processing technique. Further, the characteristics of blast cells were extracted using 4-moment statistical features and artificial neural networks (ANNs). It was found that the proposed method yielded a blasts cell segmentation accuracy of 97% under different lighting conditions.