Breaking Limitations of Fluorescent Microscopy with Learning Based Time-Lapse Imaging

While time-lapse imaging is desirable for recording dynamic development processes, such as cell division and cell migration, the image quality suffers from the fluorescent blending and phototoxicity. Theoretically, the confocal microscopy can collect images with high resolution if the livability is not the priority. In this work, we leverage the image quality of time-lapse images by formulating the transformation from timelapse image to single-point image as an adversarial learning task, which bypasses the lack of pairwise images for training. Compared with other non-blind image denoising algorithms, our method shows compelling performance in restoring imaging targets.