Bio-CM{superscript 2}: Distributed computational optics for cortex-widecellular imaging

Technical Summary of Bio-CM2: Distributed Computational Optics for Cortex-Wide Cellular Imaging

Problem Statement
Understanding distributed biological systems, particularly neural circuits, requires simultaneous imaging at cellular resolution across millimeter-scale fields of view (FOV). Existing miniature microscopes face fundamental trade-offs among FOV, spatial resolution, and optical complexity. Conventional single-objective designs suffer from rapid off-axis aberration degradation as FOV increases, necessitating complex optics that compromise the compact form factor required for in vivo applications. Alternative strategies, such as camera arrays, introduce significant hardware complexity and synchronization burdens, while computational imaging methods relying on optical multiplexing (e.g., light-field or lensless imaging) often mix in-focus signals with out-of-focus background, degrading contrast and signal-to-noise ratio (SNR) under in vivo conditions. Bridging cellular-resolution microscopy with mesoscopic imaging in a compact, miniature platform remains a significant challenge.

Methodology: Distributed Computational Optics
The authors introduce Bio-CM2, a computational miniature mesoscope that implements a "distributed computational optics" framework. Rather than relying on a single objective lens or optical multiplexing, Bio-CM2 distributes image formation across four coordinated optical modules arranged in a 2×2 configuration.

• Optical Architecture: The system utilizes four independent miniature plano-aspheric lenses (3 mm diameter, 3.52 mm focal length) that image partially overlapping subfields onto a single shared backside-illuminated CMOS sensor (4000×3000 pixels). This design partitions the total FOV before image formation, ensuring that each sensor pixel receives light from only one object-space location within a single channel. This approach avoids the contrast loss associated with optical multiplexing while reducing the maximum field angle for each lens, thereby alleviating off-axis aberrations without increasing optical complexity.
• Illumination: Four symmetrically positioned LEDs (470 nm) surround the lens array, utilizing the built-in hemispherical lenses of the LED packages to provide near-uniform fluorescence excitation across the entire imaging region.
• Computational Reconstruction: The raw measurements undergo a modular computational pipeline:
  1. Denoising: For in vivo recordings, the self-supervised spatiotemporal denoising framework SUPPORT is employed to handle weak, temporally correlated neuronal signals.
  2. Correction: Flat-field correction compensates for illumination nonuniformity. Geometric distortion correction maps pixel coordinates using experimentally calibrated spatially varying Point Spread Functions (PSFs) and radial distortion models.
  3. Deconvolution: Patch-wise Tikhonov deconvolution is applied using the measured PSFs to restore spatial resolution.
  4. Stitching: The corrected sub-images are registered and stitched into a seamless wide-FOV image. For curved surfaces (e.g., the cortex), image warping based on manually selected feature points compensates for parallax.

Key Results
Bio-CM2 achieves a 7.5 × 10 mm² FOV with cellular-scale spatial resolution at video rates.

• System Characterization: The system demonstrates a lateral resolution with a full-width at half-maximum (FWHM) of approximately 6 µm near the optical axis, degrading to ~22 µm at the periphery due to off-axis aberrations. The axial depth of field (DOF) is estimated at ~473 µm.
• High-Throughput Fixed Tissue Imaging: The system successfully reconstructed entire 75-µm-thick coronal mouse brain sections in a single acquisition, resolving neuronal morphology across multiple regions (neocortex, hippocampus, thalamus, etc.) with fidelity comparable to conventional 4× objectives.
• Population-Scale Imaging: Bio-CM2 captured freely behaving Caenorhabditis elegans at video rates, simultaneously resolving individual fluorescent protein aggregates within multiple worms and correlating them with locomotor behavior.
• Cortex-Wide Vascular Imaging: In head-fixed mice, the system enabled simultaneous quantification of pial arteriole vasomotion (single-vessel dynamics) and mesoscale hemodynamic functional connectivity. It resolved diameter fluctuations of individual vessels and mapped regional cerebral blood volume dynamics across the dorsal cortex.
• Cortex-Wide Calcium Imaging: Using GCaMP6s-expressing interneurons, Bio-CM2 resolved the activity of 3,474 neurons across the dorsal cortex in a single recording. The system successfully separated neuronal signals from dynamic background fluorescence and vascular signals. The extracted calcium kinetics (median time-to-peak ~200 ms, half-decay ~704 ms) were consistent with known in vivo GCaMP6s properties. Furthermore, mesoscale functional connectivity analysis revealed cortex-wide activity patterns consistent with neurovascular coupling.
• Olfactory Bulb Imaging: The platform demonstrated versatility by resolving stimulus-specific functional activity in the main olfactory bulb, distinguishing glomerular responses to different odorants.

Significance and Claims
The paper claims that Bio-CM2 establishes distributed computational optics as a scalable framework for multiscale biological imaging. By distributing optical complexity across individual modules rather than a single lens, the architecture overcomes the conventional trade-off between FOV and spatial resolution while maintaining a compact form factor suitable for in vivo use.

Key contributions include:

1. Overcoming Trade-offs: Achieving a 7.5 × 10 mm² FOV with cellular resolution without the hardware complexity of multi-camera systems or the contrast degradation of optical multiplexing.
2. Multiscale Integration: Enabling simultaneous analysis from single-cell dynamics (individual neurons or vessels) to mesoscale functional networks within a single acquisition.
3. Versatility: Demonstrating applicability across diverse biological systems (mouse brain, C. elegans) and modalities (structural, vascular, and functional calcium imaging).

The authors note that while the current implementation is head-fixed due to electronics constraints, the optical and computational framework is compatible with future lightweight sensor boards for wearable applications. They also identify future directions for multimodal integration (e.g., multicolor imaging) and algorithmic improvements using deep learning for deconvolution and background removal, which could be implemented without hardware changes.