Encoder-decoder deep learning network for simultaneous reconstruction of fluorescence yield and lifetime...

Area of Science:

• Biomedical Optics
• Medical Imaging
• Fluorescence Molecular Tomography

Background:

• Time-domain fluorescence molecular tomography in reflective geometry (TD-rFMT) offers deep tissue imaging beyond conventional limits.
• Reconstructing accurate fluorescence distribution, yield, and lifetime remains a challenge.

Purpose of the Study:

• To develop and validate an end-to-end encoder-decoder network for enhanced TD-rFMT reconstruction.
• To improve spatial resolution, positioning accuracy, and lifetime accuracy in TD-rFMT.

Main Methods:

• An end-to-end encoder-decoder deep learning network was designed for TD-rFMT.
• The network directly reconstructs fluorescence yield and lifetime from time-resolved signals.
• Self-supervised and supervised joint training utilized simulated data with added noise and a custom loss function.

Main Results:

• The proposed network significantly improved spatial resolution compared to existing TD-rFMT methods.
• Enhanced positioning accuracy and accuracy of fluorescence lifetime values were demonstrated.
• Simulations and phantom experiments validated the network's performance.

Conclusions:

• The developed deep learning approach substantially enhances TD-rFMT reconstruction performance.
• This method holds promise for more accurate and detailed molecular imaging in deep tissues.
• The network effectively addresses key limitations in current TD-rFMT techniques.