Lifetime Determination Algorithms for Time-Domain Fluorescence Lifetime Imaging: A Review

Abstract

Fluorescence lifetime imaging (FLIM) is powerful for monitoring cellular microenvironments, protein conformational changes, and protein interactions. It can facilitate metabolism research, drug screening, DNA sequencing, and cancer diagnosis. Lifetime determination algorithms (LDAs) adopted in FLIM analysis can influence biological interpretations and clinical diagnoses. Herein, we discuss the commonly used and advanced time-domain LDAs classified in fitting and non-fitting categories. The concept and explicit mathematical forms of LDAs are reviewed. The output lifetime parameter types are discussed, including lifetime components, average lifetimes, and graphic representation. We compare their performances, identify trends, and provide suggestions for end users in terms of multi-exponential decay unmixing ability, lifetime estimation precision, and processing speed.

A Novel Fit-Flexible Fluorescence Soft Imager: Tri-Sensing of Intensity, Fall-Time, and Life Profile

“…To select an optimal curve describing the best data trend, various badness-of-fit (BoF) or goodness-of-fit (GoF) objective functions may be employed [62]. Generally, BoF criteria such as the two-sample Kolmogorov-Smirnov (K-S) difference [63], Kullback-Leibler (K-L) divergence [64], chi-square [13], mean squared error (MSE) [13], normalised root mean square error (NRMSE) [65] and symmetric mean absolute percentage error (SMAPE) [66], or GoF metrics such as the correlation coefficient (CC) [64] can be used.…”

“…Our proposed Misfit-percent acts as a specific type of l 1 -norm and remains outlier-robust in comparison to l 2 -norm counterparts of χ 2 and NRMSE. Although the chi-square is known as a gold standard metric in the literature of FLIM [13,62], its performance is inappropriate, due to suffering from the normalisation bias. 5) Performance under different photon budgets: Here, an experiment is designed to evaluate the performance under low, mid, and high photon-count regimes.…”

Applications of Machine Learning in time-domain Fluorescence Lifetime Imaging: a Review

“…The task of determining ( 4 ) can be translated back into a parameter estimation problem in regression or time series analysis. Different approaches have been proposed for lifetime estimation in the FLIm literature, including those enumerated by a recent review of [ 33 ] published in 2022. In this section, we provide a complementary critical review focusing more on detailed technical characteristics of developed AI- and ML-based approaches.…”

“…Statistical approaches in [ 89 , 90 ] tried to jointly estimate the IRF of h [ n ] and fluorescence lifetime components from time-resolved data using iterative methods of extended Kalman filter and expectation maximisation, respectively. Unsupervised approaches have the potential to be used in the hardware realisation of FLIm systems [ 33 ], so long as they support real-time processing constraints such as existing approaches in RLD [ 74 , 81 , 82 ] or center of mass estimation [ 85 ]. However, a number of researchers have recently shifted the custom paradigms of lifetime estimation to learning-based mechanisms to specifically profit from the past decade of progress in DL methods, leading to the second supervised category [ 77 – 80 , 93 – 96 ].…”

“…In the literature of FLIm, several valuable review papers explain the underlying physical processes and applications of FLIm exist [ 9 , 13 , 27 – 32 ]. Recently, a few researches mentioned some applications of ML to FLIm [ 24 , 33 – 35 ]. We aim to complement those studies by providing an up-to-date, detailed and structured review exclusively focused on FLIm.…”