Artifact-guided restoration improves visualization of cellular ultrastructure in cryo-FIB-SEM volumes
Artifact-guided restoration improves visualization of cellular ultrastructure in cryo-FIB-SEM volumes
Zhang, N.; Joubert, L.-M.; Wu, G.-H.; Su, G.-C.; Liu, P.; Mishra, M.; Li, P.; Sanz-Ros, J.; Einav, S.; Wang, B.; Cobos, I.; Xu, M.; Chen, M.
AbstractCryogenic focused ion beam scanning electron microscopy (cryo-FIB-SEM) enables volumetric imaging of vitrified specimens, but charging artifacts, high noise and weak membrane contrast hinder the interpretation of biological features. Here we present a two-stage deep-learning-based framework that translates cryo-FIB-SEM images toward the domain of conventional room-temperature FIB-SEM (RT-FIB-SEM), suppressing large-scale charging artifacts and enhancing local membrane signal. Held-out synthetic tests show improved agreement with constructed reference targets, while results on four experimental cryo-FIB-SEM datasets show reduced visible artifacts and clearer membrane features. Processed cryo-FIB-SEM volumes from our two-stage model can be directly analyzed by tools designed for RT-FIB-SEM, and an existing pretrained segmentation model is applied to our model output, yielding extensive predictions of membrane structures. This post-acquisition approach improves the visual interpretability and downstream analysis capabilities of cryo-FIB-SEM data without modifying imaging hardware.