Multiplexing and Sensing with Fluorescence Lifetime Imaging Microscopy Empowered by Phasor U-Net.

Area of Science:

• Biophysics
• Materials Science
• Microscopy

Background:

• Fluorescence lifetime imaging microscopy (FLIM) is vital for materials and life sciences but suffers from accuracy issues due to limited photon counts and slow processing.
• Current FLIM methods face challenges in precise lifetime estimation and handling large datasets, hindering advanced applications.

Purpose of the Study:

• To introduce Phasor U-Net, a novel deep learning approach for rapid and accurate FLIM data processing.
• To improve lifetime estimation accuracy and reduce data processing time in FLIM.
• To demonstrate the method's utility in multiplexed imaging and material characterization.

Main Methods:

• Developed Phasor U-Net, a deep learning model featuring two U-Net subnetworks for denoising and deconvolution.
• Trained the model exclusively on computer-generated datasets, eliminating the need for extensive experimental data.
• Applied the method to analyze FLIM data, including multiplexed imaging of biological samples and quantum dot characterization.

Main Results:

• Phasor U-Net achieved a 1.5-8-fold reduction in modified Kullback-Leibler divergence on phasor plots compared to direct phasor analysis.
• The method reduced the mean absolute error of lifetime images by 1.18-4.41-fold.
• Successfully demonstrated multiplexed imaging of mouse small intestine samples and improved quantum dot size estimation using lifetime information.

Conclusions:

• Phasor U-Net offers a significant advancement in FLIM, providing faster and more accurate lifetime estimations.
• The method's ability to be trained on synthetic data makes it broadly applicable and reduces experimental burden.
• This deep learning approach paves the way for new fundamental research utilizing FLIM in diverse scientific fields.