We established a volumetric trans-scale imaging system with an ultra-large field-of-view (FOV) that enables simultaneous observation of millions of cellular dynamics in centimeter-wide three-dimensional (3D) tissues and embryos. Using a custom-made giant lens system with a magnification of ×2 and a numerical aperture (NA) of 0.25, and a CMOS camera with more than 100 megapixels, we built a trans-scale scope AMATERAS-2, and realized fluorescence imaging with a transverse spatial resolution of approximately 1.1 µm across an FOV of approximately 1.5×1.0 cm². The 3D resolving capability was realized through a combination of optical and computational sectioning techniques tailored for our low-power imaging system. We applied the imaging technique to 1.2 cm-wide section of mouse brain, and successfully observed various regions of the brain with sub-cellular resolution in a single FOV. We also performed time-lapse imaging of a 1-cm-wide vascular network during quail embryo development for over 24 hr, visualizing the movement of over 4.0×10⁵ vascular endothelial cells and quantitatively analyzing their dynamics. Our results demonstrate the potential of this technique in accelerating production of comprehensive reference maps of all cells in organisms and tissues, which contributes to understanding developmental processes, brain functions, and pathogenesis of disease, as well as high-throughput quality check of tissues used for transplantation medicine.

Decoders for brain-computer interfaces (BCIs) assume constraints on neural activity, chosen to reflect scientific beliefs while yielding tractable computations. Recent scientific advances suggest that the true constraints on neural activity, especially its geometry, may be quite different from those assumed by most decoders. We designed a decoder, MINT, to embrace statistical constraints that are potentially more appropriate. If those constraints are accurate, MINT should outperform standard methods that explicitly make different assumptions. Additionally, MINT should be competitive with expressive machine learning methods that can implicitly learn constraints from data. MINT performed well across tasks, suggesting its assumptions are well-matched to the data. MINT outperformed other interpretable methods in every comparison we made. MINT outperformed expressive machine learning methods in 37 of 42 comparisons. MINT’s computations are simple, scale favorably with increasing neuron counts, and yield interpretable quantities such as data likelihoods. MINT’s performance and simplicity suggest it may be a strong candidate for many BCI applications.