FLIM assessment of cells heterogeneity: machine learning-based approach

Abstract

Fluorescence lifetime imaging (FLIM) technique is extensively used for label-free analysis of cells heterogeneity and cells response to treatment with external agents. The parameters of fluorescence decay curves are used to assess the presence of subpopulations of cells and as indicators of metabolic alterations. For instance, the shift of the average fluorescence lifetime distribution can be used as a marker of cancer cells response to chemotherapy. Two approaches can be used to analyse the FLIM data. The first one requires fitting the fluorescence decay curves pixel be pixel for the whole image and further assessment of distribution of fluorescence decay parameters. The second approach is based on the analysis of distributions of FLIM parameters of single cells. Both approaches are used in the FLIM literature, however, the question of sensitivity of whole image and single cells fluorescence parameters distributions to alteration of metabolic state was not investigated. In this work, we present the first numerical analysis of sensitivity of FLIM parameters, calculated using different algorithms, to changes in the metabolic state of the cells. We first demonstrate when two subpopulations of cells with different parameters can be detected, in other words, when can one observe bimodality in fluorescence lifetime distributions. Secondly, we present experimental results on treatment of cancer cells with chemotherapy agents and demonstrate that different FLIM processing algorithms yield different sensitivity of analysis. Namely, the most important outcome of the present work is that detection of cells subpopulations can not be performed without image segmentation and analysis of fluorescence decay over single cells.
The studies are supported by the Russian Science Foundation (grant #20-65-46018).

Speaker

Gayer Alexey
Department of Physics, M.V. Lomonosov Moscow State University,
Russia

Discussion

Ask question