Deep learning-based temporal deconvolution for photon time-of-flight distribution retrieval.

Area of Science:

• Biomedical Optics
• Computational Imaging
• Machine Learning in Science

Background:

• Time-of-flight (ToF) acquisition is crucial for biomedical applications.
• Existing methods for deconvolution of instrument response functions (IRFs) are computationally intensive and require regularization.
• Accurate deconvolution is essential for interpreting time-resolved experimental data.

Purpose of the Study:

• To introduce a novel deep learning model for deconvolution in fluorescence lifetime imaging (FLI).
• To retrieve the true photon ToF distribution by addressing distortions from the IRF.
• To provide a computationally efficient and accurate alternative to traditional deconvolution techniques.

Main Methods:

• Development and training of a deep learning model using simulated FLI data.
• Validation of the model with simulated data to assess its ability to recover true ToF distributions.
• Experimental validation using in vitro time-resolved imaging modalities with varying IRFs.
• Further validation through in vivo preclinical studies.

Main Results:

• The deep learning model successfully performed deconvolution on simulated FLI data.
• Robust performance was demonstrated across different in vitro imaging modalities with diverse IRFs.
• The model proved effective in in vivo preclinical investigations.
• The proposed method offers accurate retrieval of photon ToF distributions.

Conclusions:

• Deep learning provides a flexible and accurate approach for deconvolution in time-resolved FLI.
• The developed model simplifies the process, overcoming limitations of traditional methods.
• This technique shows significant potential for advancing diffuse optical imaging and FLI applications.