Highly multiplexed spectral FLIM via physics informed data analysis

Spectral fluorescence lifetime imaging (S-FLIM) allows for the simultaneous deconvolution of signal from multiple fluorophore species by leveraging both spectral and lifetime information. However, existing analyses still face multiple difficulties in decoding information collected from typical S-FLIM experiments. These include: using information from pre-calibrated spectra in environments that may differ from the cellular context in which S-FLIM experiments are performed; limitations in the ability to deconvolute species due to overlapping spectra; high photon budget requirements, typically about a hundred photons per pixel per species. Yet information on the spectra themselves are already encoded in the data and do not require pre-calibration. What is more, efficient photon-by-photon analyses are possible reducing both the required photon budget and making it possible to use larger budgets in order to discriminate small differences in spectra to resolve spatially co-localized fluorophore species. To achieve this, we propose a Bayesian S-FLIM framework capable of simultaneously learning spectra and lifetimes photon-by-photon ultimately using limited photon counts and being highly data efficient. We demonstrate the proposed framework using a range of synthetic and experimental data and show that it can deconvolve up to 9 species with heavily overlapped spectra.