Rapid Fluorescence Lifetime Imaging through One-Dimensional Deep Learning Optimization.

Area of Science:

• Biomedical optics
• Computational biology
• Deep learning applications

Background:

• Traditional fluorescence lifetime imaging (FLIM) analysis is computationally intensive.
• Existing methods often require complex data-fitting techniques.
• There is a need for faster and more efficient FLIM data processing.

Purpose of the Study:

• To develop a hardware-efficient deep learning approach for rapid FLIM data analysis.
• To improve the speed and reduce computational requirements of FLIM.
• To validate the performance of 1D CANNs on FLIM and related imaging techniques.

Main Methods:

• Utilized one-dimensional channel attention convolutional neural networks (1D CANNs).
• Trained 1D CANNs on raw time-correlated single-photon counting (TCSPC) data.
• Applied 1D CANNs to fluorescence lifetime fitting, phosphorescence lifetime imaging (PLIM), and STED-FLIM.

Main Results:

• Achieved high efficiency and speed in FLIM data processing with 1D CANNs.
• Demonstrated strong consistency between predicted and ground truth lifetime maps.
• Obtained prediction errors within 10% for PLIM and improved spatial resolution in STED-FLIM.

Conclusions:

• 1D CANNs offer a computationally efficient and accurate alternative for FLIM analysis.
• The method shows broad potential in various biomedical and photonics applications, including disease diagnosis and advanced imaging.
• 1D CANNs are robust across different photon count conditions.