Deep Learning Framework for 3D Fluorescence Lifetime Estimation

Fluorescence lifetime imaging has emerged as a powerful tool for quantitatively assessing the molecular environment of live tissues. While fluorescence lifetime microscopy is a maturing field, achieving effective 3D imaging in deep tissues remains challenging due to high scattering. In this study, we present a deep neural network-based approach, AUTO-FLI, which enables both 3D intensity and quantitative lifetime reconstructions at centimeter depths. Unlike conventional approaches that estimate lifetime from 2D measurements prior to reconstruction, AUTO-FLI directly recovers voxel-wise lifetime within a three-dimensional domain. The proposed method incorporates an in silico framework to generate fluorescence lifetime data for training and validation. The model is further validated using experimental data acquired on an anatomically accurate mouse-mimicking phantom. The results demonstrate accurate 3D estimates of both intensity and lifetime distributions in highly scattering media, supporting fluorescence lifetime-based molecular imaging at mesoscopic and macroscopic scales, with potential applications in pre-clinical research and fluorescence-guided surgery.