Automatic Classification of Leukemic Cells by Label-Free Light-Sheet Flow Cytometry with Machine Learning

Abstract

Accurate imaging detection at single-cell level and fast, label-free identification of leukemia cells are two major goals of research, as early diagnosis and treatment will improve the treatment of leukemia. Recently, we developed a label-free light-sheet flow cytometer to measure the two-dimensional (2D) light scattering of cells. Here we collected 2400 patterns of single cells from healthy human white blood cells, HL-60 cells (human acute myeloid leukemic cells) and K562 cells (human chronic myeloid leukemic cells, respectively. Uniform local binary pattern (LBP) was applied to extract features of the 2D patterns, which were then analyzed by the support vector machine (SVM) algorithm. The label-free classification of healthy human white blood cells with each of the two kinds of leukemia cells can reach an average accuracy rate of 98.88% and 99.38%, respectively. Furthermore, our method was applied to study healthy human white blood cells and K562 cells in cell mixtures, and the results of label-free classification of the cells agreed with the given concentration ratios. The combination of label-free light-sheet flow cytometry with machine learning is expected to realize the classification of cells from clinical leukemia patients.