Particle filter for spectral unmixing

The paper addresses the problem of identification of fluorescent molecules, or fluorophores, in biological samples obtained from time-resolved fluorescence microscopy. The contribution of this work is an algorithm, based on particle filter optimization, that solves the spectral unmixing problem for more than two fluorophores even when they present highly overlapping spectra at different wavelengths. The estimation of the proportional contributions of the elements in the sample is carried out by a Bayesian methodology. Validation is done by using synthetic mixtures with two components and ex-vivo samples from human atherosclerotic tissue.