Histogramless Time-Domain Sketched Fluorescence Lifetime Imaging

A team of researchers led by Zhenya Zang at the University of Edinburgh has published a new method for compressing fluorescence lifetime imaging (FLIM) data by a factor of up to 256x without sacrificing accuracy. Traditional FLIM relies on time-correlated single-photon counting (TCSPC) to build a histogram of photon arrival times per pixel, which consumes enormous storage and bandwidth—especially for large single-photon avalanche diode (SPAD) arrays. The team’s approach, detailed in a preprint on arXiv (2605.06532), replaces the full histogram with a sparse, non-uniform one-dimensional spline sketch. The key innovation is that knot positions in the spline are allocated based on Fisher information, concentrating sketch channels where the decay signal has the highest statistical discriminability, rather than using a uniform binning strategy.

The method was validated on both synthetic mono- and bi-exponential decay data and experimental fluorescent dye data, showing comparable performance to conventional non-linear least-squares fitting (NLSF) and Poisson maximum-likelihood estimation (MLE). Additionally, the authors demonstrated that the timestamp-to-sketch projection can be implemented directly in firmware using fixed-point lookup tables, making it feasible for real-time, high-resolution SPAD array systems constrained by data throughput. This work could enable next-generation FLIM systems to capture more pixels at higher frame rates without overwhelming data links, with applications in biological imaging, medical diagnostics, and material science.