A hybrid reconstruction of the physical model with the deep-learning that improves structured illumination microscopy

In handling raw images with low signal-to-noise (SNR) ratios, conventional algorithms of structured illumination microscopy are prone to artifacts, while deep-learning-based (DL) algorithms may lead to degradation and hallucinations. We propose a hybrid that combines the physical inversion model with a Total Deep Variation regularization. In super-resolving from low SNR images such as actin filaments, our method outperforms conventional or DL methods in suppressing artifacts and hallucinations while maintaining resolutions.