DeepFLR: a self-supervised deep learning framework for robust FLIM signal restoration under photon-starved conditions

Fluorescence lifetime imaging microscopy (FLIM) offers valuable insights into molecular states within cellular environments. However, accurate lifetime quantification under photon-starved conditions is often challenging. We introduce DeepFLR, a self-supervised deep learning framework that restores FLIM signals by exploiting spatiotemporal correlations. The restored fluorescence decay curves can then be quantified using conventional lifetime estimation algorithms, significantly reducing their minimal photons-per-pixel (PPP) requirements. Comprehensive evaluations demonstrated accurate lifetime estimation with fewer than 10 PPP across diverse biological samples. A benchmark database and a user-friendly Napari plugin will be provided. By significantly reducing photon budget, DeepFLR enables numerous low-light FLIM applications.