Application of machine learning classification algorithm for identification of cells state and line based on holographic data

Abstract

Low invasive holographic methods and machine learning classification algorithms are effective tools in cellular research. We present application of machine leaning algorithms for classification of cell states and lines basing on holographic data. HeLa, A549 and 3T3 cells in different states (living, apoptotic and necrotic) were investigated by means of digital holographic microscopy. Changes in cell state were provided by photodynamic treatment (PDTr) with different doses and were identified by standard fluorescence tests. Data on the dynamics of cells morphology was obtained by means of digital holographic microscope based on Mach–Zehnder interferometer. Processing of recorded holograms allowed for obtaining optical and morphological parameters of cells, including projected area, average phase shift and cellular dry mass. The measurement accuracy was determined by comparison with the corresponding data obtained by confocal fluorescence microscopy and comprised about 90%. An extended database of 9 holographic parameters was formed for living, apoptotic and necrotic cells of the three lines and classification algorithms were created using the obtained database. The highest classification accuracy of cell states of about 90% and of cell lines of about 92% was achieved by the support vector machines algorithm. Approbation of the developed classifiers was performed using known PDTr doses triggering apoptosis and necrosis or ensuring cell survival. The number of cells classified as living, apoptotic and necrotic was shown to be in agreement with PDTr doses leading to the corresponding death pathways. The financial support from RSF under the grant # 21-72-10044 is gratefully acknowledged.

Speaker

Anna Zhikhoreva
Ioffe Institute
Russia

Discussion

Ask question