Meta‐Learning Based Optical Aberration Compensation for Self‐Supervised Super‐Resolution Microscopy

ABSTRACT Optical aberrations affect the imaging quality of fluorescence microscopy. Traditional adaptive optics (AO) methods rely on additional hardware, such as wavefront sensors, to measure and correct optical aberrations. Although recent research on deep learning‐based aberration correction has made significant progress, most methods still rely on synthesized paired images for supervised training. This limits the scope of application when unfavorable inputs are used in multiple complex noise scenarios, requiring retraining of the model for specific aberration types. In this study, a novel meta aberration correction network (MACN) based on meta‐learning is proposed. This method eliminates the need for additional optical components and imaging steps. It connects directly to the front‐end imaging system for adaptive online training. Extensive simulation experiments demonstrate that training time and data acquisition requirements are significantly reduced while high‐precision optical aberration correction is maintained. Our proposed method enables instant adaptive optics, substantially enhancing the image resolution of fluorescence microscopy systems.