Abstract
Hyperspectral Fluorescence Lifetime Imaging allows for the simultaneous acquisition of spectrally-resolved temporal fluorescence emission decays. In turn, the rich multidimensional data set acquired enables to image simultaneously multiple fluorescent species to facilitate high-content molecular imaging for improved diagnosis. However, to enable quantitative imaging, inherent spectral overlap between the considered fluorescent probes and potential bleed-through has to be taken into account. Such task is performed typically either via spectral or lifetime unmixing, but neither both simultaneously. Herein, we present UNMIX-ME, a deep learning fluorescence un-mixing algorithm that is tasked with performing quantitative fluorophore unmixing using both spectral and temporal signatures simultaneously. UNMIX-ME was efficiently trained and validated using an in silico framework replicating the characteristics of our compressive hyperspectral fluorescent lifetime imaging platform. It was benchmarked against a conventional LSQ method for tri-exponential simulated samples. Last, UNMIX-ME performances were assessed using NIR FRET in vitro and in vivo small animal experimental data.