A deep learning method to separate fluorophores based on their fluorescence lifetime.

Abstract

Fluorescence lifetime in biological samples is a useful feature to demultiplex the fluorescence signal of spectrally overlapping fluorophores. A problem that frequently occurs when multiple fluorophores are in use is the possible spectral overlap. Therefore, selected fluorophores should be chosen to avoid this issue, consequently constraining the number and the type of fluorophores that can be used at the same time on one sample. Moreover the ability to discriminate fluorophores from temporal information does not come from their own emission spectra, but from the (fluorescent-life) time information. A well-established method for the analysis of lifetime images is the phasor approach. The phasor approach works in the frequency domain and requires no a priori information on the lifetime of fluorophores in the sample. However, the phasor method is based on the linear separation of components and does not exploit the whole information contained in the lifetime data. In this work, we propose a method to separate the contribute from two spectrally overlapping fluorophores based on fluorescence lifetime. We trained a neural network with synthetic and real images of cells stained with two spectrally overlapped fluorochromes. Future development will aim to extend our approach to our fluorescence lifetime image scanning microscopy platform, to extend the method for more than two spectrally overlapping species and to find new features in order to depend less on the training set.