Miniscope3D: optimized single-shot miniature 3D fluorescence microscopy

The study addresses the limitation of standard open-source miniature widefield fluorescence microscopes (Miniscopes), which capture only 2D information and are difficult to adapt to 3D without increasing size/weight or sacrificing resolution. The research question is how to enable high-quality, single-shot 3D fluorescence imaging in a compact, lightweight Miniscope platform. The authors propose replacing the tube lens with an optimized multifocal phase mask at the objective’s aperture stop to encode volumetric information into a single 2D measurement, reconstructable via sparsity-constrained inverse methods. This approach aims to deliver uniform resolution across a large depth range while reducing device size and weight, enabling applications such as volumetric neural imaging in freely moving animals and dynamic 3D imaging in lab environments.

The paper situates its contribution among volumetric microscopy methods. Scanning approaches (two-photon microscopy and light-sheet microscopy) offer high resolution but are challenging to miniaturize, suffer from motion artifacts and limited FoV, and require complex hardware. Miniaturized implementations exist but face trade-offs between temporal resolution and FoV and remain complex. Single-shot methods encode 3D volumes into a single 2D measurement, enabling high temporal resolution limited by camera frame rate, but require solving compressed sensing inverse problems and often suffer from non-uniform resolution over depth and larger form factors. The MiniLFM, a leading miniature single-shot 3D approach, uses a unifocal microlens array and achieves good resolution at specific depths but degrades away from the focal plane and has a smaller usable depth range. Prior analyses suggest multifocal designs could improve depth-uniformity of resolution. The authors build upon this by placing a phase mask in the Fourier plane and optimizing it for compressed sensing performance while accounting for field-varying aberrations inherent to miniature objectives.

Hardware: The system modifies a standard 2D Miniscope by removing the tube lens and placing a 55 µm-thick optimized multifocal phase mask at the aperture stop (Fourier plane) of a GRIN objective lens. The phase mask consists of an engineered pattern of multifocal microlenses with varying focal lengths to equalize resolution across depth while minimizing device size and weight. The prototype measures 17 mm in height and weighs 2.5 g. Encoding principle: Each point source emits a distinctive high-frequency multi-spot point spread function (PSF) at the sensor that varies predictably with the source’s 3D position. Placing the mask at the aperture stop approximates shift-invariant behavior and improves compactness and computational efficiency. Calibration and forward model: A sparse set of calibration PSFs (64 per depth) is acquired by scanning a 25 µm green fluorescent bead throughout the volume. These measurements are used to pre-compute an efficient forward model that captures field-varying aberrations of the miniature objective across the field of view. The forward model enables fast generation of predicted measurements for iterative reconstruction. Reconstruction: The 3D fluorescence volume is recovered from a single 2D measurement by solving a sparsity-constrained inverse problem (compressed sensing). Priors such as native sparsity and total variation are employed; reconstruction assumes samples are sparsely labeled in fluorescence. The method reconstructs up to 24.5 million voxels from a 0.3 megapixel sensor measurement. Digital refocusing is achieved via the 3D reconstruction. Design and fabrication: The authors provide design principles for optimizing phase masks for 3D imaging and fabricate the mask using two-photon polymerization on a Nanoscribe 3D printer. The system’s performance is analyzed theoretically via PSF properties and experimentally characterized. Imaging conditions and performance envelope: Demonstrated capture volume is approximately 900 × 700 × 390 µm³ at up to 40 volumes per second. Resolution is measured using fluorescent USAF targets and bead phantoms across depth; biological samples include fixed scattering and optically cleared mouse brain slices and freely swimming fluorescently stained tardigrades.

- Achieved lateral resolution of 2.76 µm uniformly over ~270 µm depth, degrading to 3.9 µm over the next ~120 µm, for a total usable depth of 390 µm. - Achieved axial resolution of 15 µm across the entire 390 µm depth range (validated via Rayleigh criterion using 4.8 µm beads), consistent with axial FWHM in 3D bead reconstructions. - Prototype dimensions and speed: 17 mm height, 2.5 g mass; captures a 900 × 700 × 390 µm³ volume at up to 40 volumes per second; reconstructs ~24.5 million voxels from a 0.3 MP image. - Two-photon verification: In 160 µm-thick 3D bead samples, Miniscope3D recovers the same features as two-photon microscopy, with slightly larger lateral spot sizes. - Biological validation: Resolves GFP-tagged neurons and dendrites (≈1 µm features) in 100 µm-thick scattering and 300 µm-thick optically cleared mouse brain slices (reconstruction volumes ≈ 790 × 617 × 210 µm³); tracks freely swimming tardigrades at up to 40 fps. - Comparison to MiniLFM: Provides 2.2× better peak lateral resolution (2.76 µm vs 6 µm best-case), ~10× larger usable depth range, ~50× more usable voxels, and smaller/lighter hardware (17 mm vs 26 mm tall; 2.5 g vs 4.7 g).

The findings demonstrate that placing an optimized multifocal phase mask at the aperture stop enables true single-shot 3D fluorescence imaging in a compact Miniscope form factor, addressing the limitations of 2D Miniscopes and improving upon MiniLFM. The approach produces nearly depth-uniform lateral resolution and 15 µm axial resolution over a large depth range, while maintaining high volumetric frame rates. By optimizing the optical encoding for compressed sensing and calibrating field-dependent aberrations, the system effectively reconstructs large volumetric datasets from undersampled 2D measurements using sparsity priors (e.g., total variation). The device’s reduced size and weight make it suitable for head-mounted applications and dynamic biological studies. While the method assumes sample sparsity and no partial occlusions and can be affected by scattering at depth, results in both scattering and cleared tissues show single-neuron resolution, indicating robustness for many biological applications. The architecture’s generality and open-source ecosystem facilitate adoption and customization.

The work introduces Miniscope3D, a miniature single-shot 3D fluorescence microscope that replaces the tube lens with an optimized multifocal phase mask at the Fourier plane, enabling compact, lightweight hardware with high, depth-uniform resolution and fast volumetric imaging. The authors provide design and fabrication guidelines for phase masks, an efficient calibration-driven forward model accounting for field-varying aberrations, and a sparsity-constrained reconstruction algorithm. Experiments on phantoms and biological samples validate performance and show significant advantages over MiniLFM in resolution, usable depth range, and device size/weight. Potential future directions include incorporating temporal, application-specific priors to further improve axial resolution and robustness, extending the model to handle partial occlusions, and exploring performance in more strongly scattering regimes and additional biological preparations.

- Assumes sample sparsity in some domain; reconstruction quality degrades as sparsity decreases. - Assumes no partial occlusions, a common limitation in 3D fluorescence recovery methods. - Susceptible to degradation from tissue scattering at greater depths, as with other single-photon microscopes. - Compared to 2D Miniscope, exhibits lower SNR and slightly reduced lateral resolution (2.76 µm vs ~2 µm) for purely 2D imaging tasks. - Requires calibration of PSFs across depth and field to model field-varying aberrations.