Сomparative analysis of structural features of multidimensional data of peripheral blood in chronic lymphocytic leukemia and non-hodgkin B-lymphoma

Abstract

Relevance. Correct identification of the lymphoproliferative diseases chronic lymphocytic leukemia (CLL) and non-Hodgkin B-cell lymphomas (NHL) is a key screening step that requires high accuracy and interpretability. Therefore, the development of modern methods for the differential diagnosis of NHL and CLL is still relevant. Non-invasive methods of differential diagnosis based on research followed by mathematical analysis using artificial neural networks (ANN) will make it possible to differentiate these diseases with high accuracy only based on blood data, without using material from other organs and tissues.

The purpose of this study is to conduct a multivariate analysis of the composition of subpopulations of lymphocytes and tumor cells in peripheral blood and evaluate the relationships in the “immunity – tumor growth” system by introducing artificial neural network (ANN) methods. This allows us to identify structural features that distinguish NHL from CLL.

Methods. Data analysis was carried out using an artificial neural network – self-organizing Kohonen maps (SOM). This ANN allows one to reveal the structure in multidimensional data by projecting multidimensional images into a reduced-dimensional space (2- or 3-dimensional), which does not requires the adoption of any a priori hypotheses about the structure of the data.

Results. Artificial neural networks (ANN) were used to construct and search for differences between multidimensional images of CLL disease and NHL. The state of immunity and tumor cells was compared according to peripheral blood data of patients with B-cell NHL (352 patients) with data from CLL (315 patients). The structures reflecting the differences between the states “NHL – CLL” have been identified. The nature of the distributions and values of immunity and tumor growth indicators in multidimensional space allows us to distinguish with high accuracy the conditions of CLL – NHL, determine the stage of tumor development and subsequently select the correct treatment tactics.

Conclusion. The differences between NHL and CLL, according to multivariate analysis, create the basis for a non-invasive method for automatic diagnosis of NHL and CLL pathologies.