Feng Tian, Weijian Yang
Mask-based integrated microscopy represents a powerful computational imaging paradigm, enabling 3D imaging over a large field of view within a thin device footprint. This architecture leverages a thin optical mask to encode object information onto a camera sensor placed in close proximity. A major challenge in this field is to maintain high reconstruction quality amid low signal-to-background ratios, while simultaneously reducing computational requirements to allow for larger reconstruction volumes. Here, we describe a significant breakthrough in this area: DeepInMiniscope, a system featuring a highly scalable reconstruction algorithm for high-quality imaging over large 3D volumes. We detail the joint development of the hardware and reconstruction algorithms that enable this high-quality, efficient, and robust imaging framework, and present its capability to record neural activity in the mouse brain with near-cellular resolution. DeepInMiniscope represents a key step towards large-scale, 3D functional recording in freely-behaving animals.