Fig. 1: flimGANE (fluorescence lifetime imaging based on Generative Adversarial Network Estimation) is a rapid and accurate method to generate fluorescence lifetime microscopy (FLIM) images.

a–c Schematic of the deep learning framework for flimGANE architecture. a Generator (G) is used to transform the acquired decay curve into an artificial high-photon-count decay. It comprises two CNN blocks, each of which is made up of one convolutional layer followed by an average pooling layer of stride four. The CNN section is followed by a flatten layer. Then a multi-task layer converts data into virtual lifetime parameters, followed by two fully connected layers. Skip connection is used to pass data between layers of the same level. b Discriminator (D) consists of four fully connected layers. c Estimator (E) comprises a partially and a fully connected layers followed by a multi-task layer to map the high-photon-count decay curve into lifetime parameters. d Comparison of FLIM images generated by different methods (n = 1340 pixels). e Comparison of squared errors by different methods. Under the ultra-low-photon-count condition (50 counts per pixel), flimGANE provides the best estimation. Under the low-photon-count condition (100 counts), TD_LSE and DFD_LSE fail to generate accurate FLIM images. Under the high-photon-count condition (1500 counts), all FLIM images match well with the ground truth. TD_LSE result exhibits no significant difference from DFD_LSE result under the low-photon-count condition based on the two-tailed t test. f flimGANE successfully characterizes the apparent lifetimes of the two-dye mixtures (stock solution: 3 μM Cy5-NHS ester and 7 μM Atto633 in DI water). The mean values obtained from Gaussian fitting are indicated as white solid circles.