New approaches to the diagnosis of Non-Hodgkin's B-cell lymphomas and Chronic Lymphocytic Leukemia by clusters of differentiation of normal and tumor lymphocytes in peripheral blood

Abstract

Background. Currently, machine learning and artificial intelligence technologies are rapidly developing and being implemented in scientific research, which significantly simplifies the study and use applied developments in medicine. The development of a method for automatic classification of Non-Hodgkin's B-cell lymphomas (NHL) and Chronic Lymphocytic Leukemia (CLL) in comparison with the norm (N) will allow us to analyze the differences in multidimensional relationships in the "immunity - tumor growth" system and will create the ability to classify these diseases based on blood data without using materials from other organs and tissues for diagnosis.

Purpose of this study - is to create an automated classification of NHL and CLL in comparison with the norm and to evaluate the quality of diagnostics based on differences in the indicators of multidimensional images of immunocompetent and tumor cells in peripheral blood.

Methodology. A comparative study of immunity and tumor cells in the peripheral blood of 185 healthy individuals, 352 patients with B-cell NHL, and 315 patients with CLL was conducted. The machine processing method classifying data by constructing a decision tree (DT) was used to compare the indicators of multidimensional images formed by lymphocyte subpopulations and tumor cells in the peripheral blood. Three classifiers of states were constructed: NHL versus normal (N–NHL), CLL versus normal (N–CLL), and NHL versus CLL (NHL–CLL). The quality of the classifiers was determined by the error’s matrix and ROC curve criteria.

Results. The classifiers of values and distribution patterns of immunity and tumor cell indices constructed by the DT method differentiate between the states of N–NHL, N–CLL and NHL–CLL and determine the classification algorithm. The criteria of the error’s matrix and ROC curve methods confirmed the high quality of the classifiers.

Conclusion. Using the machine learning method, a DT was constructed that classifies NHL and CLL, which allows for the implementation of alternative diagnostics of NHL and CLL using only data on the presence of immune and tumor lymphocytes in the peripheral blood.