Correlative expansion imaging-guided deep learning to enable high-throughput in vivo super-resolution microscopy

In vivo microscopy is a crucial technique for visualizing and understanding the complex dynamics of live organs, including the brain. While fluorescence microscopy methods, such as two-photon microscopy, enable cellular-resolution imaging of live tissues, their inherent resolution limits constrain deeper insights. In this work, we introduce an integrated in vivo– expansion correlative imaging pipeline coupled with an unsupervised deep learning algorithm to achieve super-resolution imaging using two-photon microscopy. Our approach leverages a registration-free super-resolution framework based on CycleGAN, incorporating physics-aware rescaling to address anisotropic scaling discrepancies between two-photon and expansion microscopy datasets. High-resolution ground truth images were obtained via expansion microscopy on the same tissues previously imaged in vivo. Using this ground truth, our deep learning model successfully enhanced the resolution of three-dimensional in vivo images, enabling high-fidelity visualization of subcellular structures.