Overcoming photon and spatiotemporal sparsity in fluorescence lifetime imaging with SparseFLIM

Area of Science:

• Biomedical imaging
• Optical microscopy
• Deep learning applications

Background:

• Fluorescence lifetime imaging microscopy (FLIM) offers valuable biochemical microenvironment insights for biomedical applications.
• Conventional FLIM acquisition is slow due to photon counting, limiting its speed and practical use.

Purpose of the Study:

• To introduce SparseFLIM, a novel deep learning paradigm for high-fidelity FLIM reconstruction from sparse photon data.
• To overcome the speed limitations of traditional FLIM techniques.

Main Methods:

• Development of a coupled bidirectional propagation network for photon enrichment and spatial-temporal information recovery.
• Reconstruction of FLIM data from sparse photon measurements.

Main Results:

• Achieved over tenfold photon enrichment, significantly boosting signal-to-noise ratio and lifetime accuracy.
• Successfully reconstructed spatially and temporally undersampled FLIM at full resolution.
• Demonstrated strong generalization across multispectral and in vivo endoscopic FLIM modalities.

Conclusions:

• SparseFLIM enables rapid, high-fidelity FLIM imaging by intelligently processing sparse photon data.
• Deep learning offers a powerful solution to enhance FLIM and overcome acquisition speed limitations.
• This method broadens the applicability of FLIM in demanding biomedical imaging scenarios.