High compression deep learning based single-pixel hyperspectral macroscopic fluorescence lifetime imaging in vivo.

Area of Science:

• Biomedical Optics
• Deep Learning in Imaging
• Fluorescence Lifetime Imaging Microscopy (FLIM)

Background:

• Single pixel imaging (SPI) offers high-dimensional data acquisition in photon-starved biological settings.
• Existing SPI frameworks suffer from slow acquisition speeds and limited pixel resolution.
• In vivo applications require faster imaging and higher resolution than currently achievable with SPI.

Purpose of the Study:

• To introduce NetFLICS-CR, a convolutional neural network for compressed sensing fluorescence lifetime imaging.
• To enhance resolution, acquisition speed, and processing speed for in vivo applications.
• To enable high-compression imaging without the need for experimental training datasets.

Main Methods:

• Developed a convolutional neural network (NetFLICS-CR) integrating compressed sensing for fluorescence lifetime imaging.
• Achieved high compression ratios (up to 99%) reducing measurements to 1% of requirements.
• Reconstructed intensity and lifetime data at 128x128 pixel resolution across 16 spectral channels.

Main Results:

• Reduced acquisition time from ~2.5 hours (50% compression) to ~3 minutes (99% compression).
• Demonstrated in silico, in vitro, and in vivo (mice) imaging capabilities.
• Successfully monitored receptor-ligand interactions and intracellular drug delivery (Trastuzumab) in tumors.

Conclusions:

• NetFLICS-CR significantly accelerates data acquisition and improves resolution in fluorescence lifetime imaging.
• The method facilitates in vivo monitoring of lifetime properties and drug uptake.
• This advancement enables the translation of single pixel macroscopic fluorescence lifetime imaging (SP-MFLI) for clinical applications.