Fig. 4: Temporal and photon sparsity enhancement by SparseFLIM.
From: Overcoming photon and spatiotemporal sparsity in fluorescence lifetime imaging with SparseFLIM
a Network architecture of TU SparseFLIM. \({{{{\boldsymbol{U}}}}}_{{{{\boldsymbol{t}}}}}\), temporal upsampling module. Dark boxes represent the collected frames, while white boxes represent the uncollected frames. b Images of RBCs and dermis, with the temporally padded frames reconstructed by the network outlined in green. c Orthogonal views of an RBC. The white dashed lines and arrows in the \({{{\boldsymbol{y}}}}\)-\({{{\boldsymbol{t}}}}\) views indicate selected \({{{\boldsymbol{x}}}}\)-\({{{\boldsymbol{y}}}}\) frames displayed at the ends. The column graph at the bottom right illustrates the SNR improvement, with mean ± SD. d Comparison of the input and output images of skin tissue, with the photon-rich image serving as the reference. e Violin plot demonstrates changes in lifetime correlation in skin data between the input (bicubic temporal upsampling) and TU SparseFLIM reconstruction. n = 500,197-lifetime traces. f Fluorescence decay of the location indicated by the cross in (d). Two-tailed Wilcoxon matched-pairs signed rank tests were applied between the input and output in (e). Scale bars, 150 μm in (b), 100 μm in (d).
