Understanding cellular events and their connections within whole organs and organisms is a fundamental quest in biology. Three-dimensional (3D) high-resolution imaging across mesoscale volumes is crucial for extracting cellular profiles of different physiological functions. However, conventional 3D light microscopy methods face challenges due to limited optical throughput and imaging depth. Strategies like 3D tile stitching with tissue sectioning have been employed, but these methods, such as tiling confocal microscopy and sequential two-photon tomography (STPT), suffer from lengthy acquisition times. Light-sheet microscopy, integrated with tissue clearing, offers an alternative, nondestructive approach, but even this faces limitations with Gaussian or Bessel light sheets. Gaussian light sheets, while useful for large samples, have a tradeoff between sheet thickness and uniformity, affecting the resolution of fine structures. Axially swept light-sheet microscopy (ASLM) improves axial resolution but leads to significant photobleaching. Bessel-type light-sheet microscopy generates non-diverging sheets, but often utilizes high NA detection objectives with small depth of focus, limiting its large-scale applications. Compressed sensing (CS) has been suggested as a way to improve throughput, but conventional CS methods are sensitive to signal characteristics and struggle with large-scale 3D images where sparsity and dynamic range vary greatly. This paper addresses these challenges by combining large-FOV scanning Bessel light-sheet microscopy with a 3D content-aware compressed-sensing (CACS) computation procedure.

The authors review existing methods for 3D imaging of large organs and organisms, highlighting the trade-offs between resolution, speed, and imaging depth. They discuss confocal microscopy, two-photon excitation microscopy (TPEM), and selective plane illumination microscopy (SPIM) and their limitations in achieving both high resolution and speed for large-scale imaging. They examine previous attempts to utilize compressed sensing (CS) in microscopy, noting challenges in dealing with variations in signal characteristics across large volumes. Existing approaches using nonlocal total variation and Laplacian scale mixture models show limited success in restoring fine details in decimated signals. The literature review establishes the need for a new approach that combines advanced illumination techniques with a robust CS method.

The authors developed a zoomable, line-synchronized, CACS Bessel plane illumination microscopy system. The system utilizes a dual-side confocally-scanned Bessel light-sheet to illuminate regions of interest, providing uniform optical sectioning with thin axial confinement. A key innovation is the line synchronization of the sweeping Bessel beam with the camera's rolling active pixel lines to block the influence of side lobes and reduce background noise. The content-aware compressed sensing (CACS) procedure adapts to the varying signal characteristics in different sample regions, further improving image quality. The CACS computation is performed in the Fourier domain, where signals are more sparsely represented. A compressed matrix A is obtained from the system's point spread function (PSF). The image stack Y is divided into small volumes, and their entropy parameter β and density parameter α are calculated. A regularization factor λ, varying with the image content, is determined and applied to the L1-norm of the Fourier representation of the high-resolution image x. This process balances the iterative signal reconstruction, avoiding both overfitting and underfitting. The high-resolution image tiles are recovered, transformed back to spatial volumes, and stitched together to create the final high-resolution image. The authors detail the optical setup, including the use of axicons, galvanometer mirrors, sCMOS cameras, and motorized stages for precise control and high-speed acquisition. The CACS computational procedure is meticulously described, including the iterative optimization method and multi-GPU parallelization for efficient processing.

The authors demonstrate superior axial resolution and contrast compared to epi-illumination and Gaussian plane illumination. The CACS procedure significantly improves resolution and SNR compared to conventional CS and deconvolution. Comparative analysis with confocal, TPE, and SPIM methods reveals superior axial and lateral resolution of the CACS Bessel sheet mode. The CACS Bessel sheet mode achieves a much higher throughput than other methods. Photobleaching experiments show significantly less photobleaching with the Bessel sheet and CACS methods compared to confocal and TPE microscopy. Whole-brain imaging of mouse brain (Thy1-GFP-M) is accomplished with isotropic subcellular resolution (0.5 µm iso-voxel size) in approximately 10 minutes. This allows for various system-level analyses including neuron tracing and segmentation. The accuracy of the CACS reconstruction is validated by comparing it to high-resolution images obtained with a higher magnification Bessel sheet. Three-dimensional mapping of cell nuclei (propidium iodide-labeled) in half a mouse brain is performed, enabling accurate cell counting and density quantification in different brain regions. Dual-color 3D imaging of mouse muscles (Thy1-YFP and α-BTX) is demonstrated, allowing quantitative analysis of neuromuscular junction (NMJ) occupancy.

The findings demonstrate the effectiveness of combining CACS with long-range Bessel plane illumination for rapid, high-resolution 3D imaging of large samples. This approach overcomes the limitations of anisotropic resolution and low throughput found in existing techniques. The CACS method's ability to handle varying signal characteristics across large-scale images is crucial for achieving high-quality reconstructions. The successful applications to whole-brain imaging, cell counting, and NMJ analysis highlight the versatility and potential impact of this technique for various biological studies. The ability to obtain whole-brain imaging data in minutes, compared to hours or days with existing methods, represents a significant advancement, opening up new possibilities for large-scale anatomical mapping and quantitative analysis. The increased throughput achieved by this technique expands the range of experimental possibilities for neuroscience, histology, and pathology research.

This research presents a novel light-sheet microscopy technique that combines dual-side confocally-scanned Bessel light-sheet illumination with a content-aware compressed-sensing (CACS) computation pipeline. This results in a significant improvement in the speed and resolution of 3D whole-organ imaging. The method is demonstrated to achieve minute-timescale high-resolution mapping of entire macro-scale organs. This advancement has the potential to significantly impact various fields of biological research, enabling large-scale studies that were previously not feasible. Future research could focus on further optimizing the CACS algorithm, exploring different tissue clearing techniques, and expanding the application of this method to other biological systems.

While the CACS method significantly reduces acquisition time and improves resolution, there are limitations to consider. The method relies on the sample being sufficiently cleared and labeled to allow for efficient light penetration. The computational processing time, while reduced compared to traditional stitching techniques, might still pose a limitation for extremely large datasets. The accuracy of the CACS reconstruction depends on the accuracy of the PSF estimation. The effectiveness of the CACS algorithm might vary depending on the specific tissue type and labeling density. Further optimization of the algorithm and hardware could improve performance for even larger and more complex samples.