Fluorescence image deconvolution microscopy via generative adversarial learning (FluoGAN)

Abstract

We propose FluoGAN, an unsupervised hybrid approach combining the physical modelling of fluorescence microscopy timelapse acquisitions with a generative adversarial learning procedure for the problem of image deconvolution. Differently from standard approaches combining a least-square data term based on one (long-time exposure) image with sparsity-promoting regularisation terms, FluoGAN relies on a data term being the distributional distance between the fluctuating observed timelapse (short-time exposure images) and the generative model. Such distance is computed by adversarial training of two competing architectures: a physics-inspired {\em generator} simulating the fluctuating behaviour as a Poisson process of the observed images combined with blur and undersampling, and a standard convolutional {\em discriminator} network. FluoGAN is a fully unsupervised approach requiring only a fluctuating sequence of blurred, undersampled and noisy images of the sample of interest as input. It can be complemented with prior knowledge on the desired solution such as sparsity, non-negativity etc. After having described the main ideas behind FluoGAN, we formulate the corresponding optimisation problem and report several results on simulated and real phantoms used by microscopy engineers to quantitatively assess spatial resolution. The comparison of FluoGAN with state-of-the-art methodologies shows improved resolution, allowing for high-precision reconstructions of fine structures in challenging real {\em Ostreopsis cf Ovata} data. The FluoGAN code is available at: https://github.com/cmayeul/FluoGAN.