Non-negative Matrix Factorization: A blind sources separation method to unmix fluorescence spectra

Fluorescent imaging in diffusive media is an emerging modality for medical applications. Here, we use spectrally resolved measurements in order to separate several fluorescence sources. It is useful to filter parasite signal (such as the intrinsic biological tissues fluorescence, called autofluorescence) or enable the multiplexing of several fluorophores. A spectroscopic approach, based on the Non-negative Matrix Factorization (NMF) method, is explored to unmix overlapping spectra and thus isolate the specific fluorescence signals. This blind source separation method treats specific fluorescences (due to the various fluorophores) as different sources to separate; it only needs initial spectra, updated over iterations thanks to the classical gradient descent method. Fluorescence contributions of Alexa750 and ICG have been satisfactorily isolated on experimental data.