Enhancing Fluorescence Lifetime Imaging With Differential Transformer

Area of Science:

• Biomedical Optics
• Medical Imaging
• Deep Learning

Background:

• Conventional fluorescence lifetime imaging (FLI) relies on accurate instrument response function (IRF) and decay data.
• IRF variations at macroscopic scales hinder precise lifetime parameter estimation.
• Current deep learning methods struggle with complex biological and in vivo imaging data.

Purpose of the Study:

• To develop a robust deep learning framework, MFliNet, for accurate multi-exponential decay parameter estimation in FLI.
• To address topographical distortions affecting IRF and photon arrival distributions.
• To enable reliable, real-time macroscopic FLI for challenging applications.

Main Methods:

• Introduced MFliNet, a deep learning framework utilizing a Differential Transformer encoder-decoder architecture.
• MFliNet jointly processes temporal fluorescence decay and IRF inputs.
• The model incorporates photon time-of-flight deconvolution principles.

Main Results:

• MFliNet effectively corrects topographical distortions in photon arrival distributions.
• Validated with tissue-mimicking phantoms and preclinical tumor models.
• Demonstrated exceptional robustness and precision in macroscopic FLI.

Conclusions:

• MFliNet provides accurate multi-exponential decay parameter estimation.
• The framework enables reliable, real-time macroscopic FLI.
• MFliNet is suitable for complex biological and intraoperative imaging scenarios.