Fluorescence lifetime (FLI) parameter estimation of a fluorescence inclusion inside a tissue remains challenging without due correction from Instrument Response function (IRF). Mathematical models, non-linear least-square-fit (‘reconvolution’), center-of-mass (CMM), and Phasor plot methods use IRF correction, however, recent machine learning (ML) models omit correction learning from IRF and often fails in in-vivo samples. Here, we use a transformer-ML model (MFLI-NET) which also takes temporal-point spread function (TPSF) and pixelwise IRF inputs to provide the offset correction due to depth. The MFLI-NET model showed high accuracy and robustness when tested with 1- and 2- exponential in vitro and in-vivo fluorescence samples.
|