henhaobio
Sign in

bioRxivBioengineeringPreprint

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

Guorong HuiD, Qilin DengiD, Tianrui Qi, Zhixiong CheniD, Bradley C. RauscheriD, Nathan X. ChaiiD, Daria BogatovaiD, Bethany Weinberg, Jesse SmithiD, Ian G. DavisoniD, Martin ThunemanniD, Anna DevoriD, Lei TianiD

PDFDOI
In-depth full-text digest available · Jump to digest ↓

Abstract

Understanding distributed biological systems, particularly neural circuits, requires simultaneous cellular-resolution imaging across millimeter-scale fields of view (FOV). Existing miniature microscopes remain fundamentally constrained by trade-offs among FOV, spatial resolution, and optical complexity, limiting their ability to bridge cellular microscopy with cortex-scale imaging. Here we introduce distributed computational optics, a framework that distributes image formation across coordinated optical modules and computationally integrates their measurements into a unified image. We realize this framework in Bio-CM2, a computational miniature mesoscope that partitions the imaging field across four optical modules while converging their measurements onto a common image sensor. This architecture overcomes the aberration-scaling limitations of conventional miniature optics while avoiding the hardware complexity of multi-camera systems and the contrast degradation associated with optical multiplexing. Bio-CM2 achieves a 7.5 x 10 mm2 FOV while enabling cellular-resolution in vivo imaging at video rates. We demonstrate its utility through two complementary imaging modalities in head-fixed mice: cortex-wide functional vascular imaging, enabling simultaneous quantification of pial arteriole vasomotion and mesoscale hemodynamic functional connectivity, and cellular-resolution calcium imaging, resolving the activity of over 3,000 neurons together with mesoscale neuronal functional connectivity. We further demonstrate the versatility of the platform through cellular-resolution imaging of entire coronal mouse brain sections, population-scale imaging of freely behaving Caenorhabditis elegans, and odor-evoked calcium imaging of the main olfactory bulb in head-fixed mice, highlighting its broad applicability across diverse biological systems and imaging modalities. By overcoming the conventional trade-off between FOV and spatial resolution in a compact miniature platform, Bio-CM2 establishes distributed computational optics as a scalable framework for multiscale biological imaging.

In-depth summary

Claude Opus 5.5 · xhigh effort · 8 Nov 2026 · from the full text of version 1 (28 Jul 2026) · CC BY 4.0

Four miniature lenses sharing one CMOS sensor, with computational stitching, give cellular-resolution imaging over 7.5 × 10 mm^2 at video rate, resolving 3,474 mouse cortical neurons.

Miniature microscopes trade field of view against resolution: single-objective designs suffer growing off-axis aberrations, camera arrays add cost and synchronization burden, and multiplexed computational optics (lensless, light-field, the earlier CM^2) lose contrast to out-of-focus fluorescence in vivo. Bio-CM^2 partitions the field before image formation across four lens modules on one sensor, so each pixel sees one object location and contrast is preserved.

Key results

  1. ShownCalibrated resolution is near-cellular at each lens axis and degrades toward module edges, the depth of field is long, and the four sub-images remain separate on the sensor up to the designed field size.

    Point spread function (PSF) major-axis full-width at half-maximum (FWHM) about 6 µm near each module axis, rising to about 22 µm at lens periphery (2.89 µm/pixel sampling, numerical aperture 0.17); axial Gaussian fit gives depth of field about 473 µm; masked target shows no sub-image overlap up to 7.5 × 10 mm^2; PSFs from a 4 µm bead on a 31 × 41 grid. · 2.2 Implementation and characterization of…; FOV characterization; PSF calibration

  2. ShownIn GCaMP6s VIP-interneuron mice imaged through a bilateral cranial window, self-supervised denoising (SUPPORT) followed by CNMF-E and quality control extracted thousands of single neurons across both hemispheres with transient kinetics typical of GCaMP6s.

    CNMF-E retained 3,474 neurons across the full field from a 3 min, 10 Hz recording; median tpeak 200.0 ms (interquartile range 168.5–236.7 ms), median t1/2 703.8 ms (609.7–851.6 ms); a two-photon maximum-intensity projection over 100–300 µm depth best matched the Bio-CM^2 image. · 2.6 Cortex-wide calcium imaging from single…; In vivo head-fixed mouse imaging with Bio-CM2; Two-photon in vivo head-fixed mouse imaging; Single-neuron calcium signal extraction

  3. ShownOne FITC-dextran recording yielded both single-arteriole diameter dynamics (vasoconstriction and vasodilation) across the dorsal cortex and cortex-wide hemodynamic functional connectivity from atlas-defined parcels.

    5 min recording at 10 Hz; relative diameter (Δd/d) power spectra show vasomotion mostly below 0.5 Hz; ΔF/F from 28 registered Allen Mouse Brain Atlas parcels; seed maps (MOs, SSp-ll, SSp-bfd, VISp) show bilateral symmetry and correlation decreasing with distance, assessed qualitatively without reported statistics. · 2.5 Cortex-wide vascular imaging from…; In vivo head-fixed mouse imaging with Bio-CM2; Vessel diameter analysis; Regional hemodynamics and neuronal activity

  4. SuggestedScan-free single acquisitions covered a whole fixed coronal mouse brain section with neuronal morphology resembling a benchtop objective image, and resolved individual polyQ aggregates in many freely moving C. elegans simultaneously.

    75-µm-thick section with GFP-labeled neurons across multiple regions including neocortex, hippocampus, thalamus, hypothalamus and amygdala; regions of interest compared visually with a 4×/0.20 NA objective (no fidelity metric reported); AM141 Q40::YFP worms imaged at video rate without immobilization. · 2.3 High-throughput imaging of fixed brain…; 2.4 Population-scale imaging of C. elegans; Fixed brain slice; Caenorhabditis elegans

  5. SuggestedA single sub-image sufficed to record odor-specific glomerular activation in the dorsal main olfactory bulb of Tbet-Cre;Ai148 mice expressing GCaMP6f, extending the platform beyond cortex.

    92 manually selected glomerular regions of interest; pentyl acetate (2%) and benzaldehyde (10%) evoked distinct ΔF/F activation maps; 20 Hz video for about 30 s with 4 s manual odor delivery; responses described qualitatively, no statistics given. · 2.7 Odor-evoked calcium imaging in the main…; In vivo head-fixed mouse imaging with Bio-CM2; Glomerular activity analysis

Methods, data and reproducibility

A 2 × 2 array of off-the-shelf plano-aspheric lenses (0.64× magnification) images four partially overlapping subfields as non-overlapping sub-images on one CMOS sensor under four-LED 470 nm excitation; a six-step pipeline (denoising, flat-field, background removal, distortion correction, patch-wise Tikhonov deconvolution with measured point spread functions, stitching) fuses them.

  • Whole 75-µm fixed coronal brain section, GFP neurons via retrograde AAV (Addgene #37825-AAVrg) injected in BNST; male C57BL/6 mice · mouse · n = single coronal section
  • Transgenic C. elegans AM141 (unc-54p::Q40::YFP) expressing polyQ-YFP in body-wall muscle; freely behaving day 1–3 adults on an agar pad · C. elegans
  • VIP-Cre;Ai162 mice (GCaMP6s in VIP interneurons) with bilateral full-dorsal-cortex glass window; cortex-wide calcium imaging, 3 min at 10 Hz · mouse · n = 3,474 neurons; same animal as vascular imaging
  • Cortex-wide vascular imaging after intravenous 2 MDa FITC-dextran, 5 min at 10 Hz, through the same chronic window · mouse · n = same animal as calcium imaging
  • Tbet-Cre;Ai148 cross (GCaMP6f in second-order main olfactory bulb neurons), thinned-skull window, two odorants · mouse · n = 92 glomerular ROIs
  • Awake two-photon z-stacks (920 nm, 5 µm steps, 0–350 µm) of the matching cortical region for depth and neuron validation · mouse
  • PSF calibration set: 4 µm fluorescent bead scanned over the 7.5 mm × 10.0 mm field at 250 µm steps · n = 31 × 41 grid positions
  • Allen Mouse Brain Atlas dorsal projection registered to the imaging field for cortical parcellation · mouse (Allen Mouse Brain Atlas) · n = 28 bilateral cortical parcels
Methods & tools
SUPPORT (self-supervised spatiotemporal denoising), BM3D (static-sample denoising), CNMF-E (neuron source extraction), Modified dark-sectioning with dark-channel prior (background removal), Patch-wise Tikhonov deconvolution with alpha blending, Radial distortion model fitted to calibrated PSF positions, Allen Mouse Brain Atlas, Bruker Ultima Investigator Plus two-photon microscope, The Imaging Source DMM 37UX226-ML CMOS sensor, Edmund Optics #16-684 plano-aspheric lens, Nikon Plan Apo λD 4×/0.20 NA reference objective, Local-maximum intensity detection (CNMF-E cross-validation)
Reproducibility gaps
No code or data repository, URL or availability statement appears in the text; the bill of materials and SUPPORT and CNMF-E settings are deferred to supplementary sections; Boston University has filed a patent application on the technology.
Limitations (6)
  • Only head-fixed mouse imaging was shown; the authors attribute this to the current imaging electronics rather than the optics and propose lighter sensor boards and flexible interconnects for a wearable version. · stated by the authors
  • The low-cost CMOS sensor has higher read noise than scientific CMOS cameras, so in vivo calcium analysis depends on SUPPORT denoising; without it CNMF-E often fails to find reliable neuronal components. · stated by the authors
  • The reconstruction was designed for robustness rather than peak quality; the authors note that patch-wise Tikhonov deconvolution and dynamic background removal for calcium imaging could be improved with deep learning methods. · stated by the authors
  • Curved cortex causes parallax between neighboring views, corrected by warp fields from manually chosen feature points; vessel-adjacent motion and hemodynamic fluctuations can create false-positive neuron components that need extra screening. · stated by the authors
  • Validation is largely qualitative and from single recordings: the brain-section comparison is visual, the neuron count lacks a two-photon match rate, and animal numbers and statistics for connectivity and odor responses are not reported. · noted in this summary
  • Resolution varies within each module (about 6 µm centrally vs about 22 µm major-axis FWHM at edges), despite claims of uniform image quality after stitching, and neuron detection yield by field position is not reported. · noted in this summary

Read this if you need cellular-resolution fluorescence imaging over centimeter-scale fields in a compact, low-cost device, for example cortex-wide neuron and vascular recordings in head-fixed mice or population imaging of small organisms.

Citations
1

Reproducibility ledger

Nobody has reported trying to reproduce this paper yet. Did you replicate an experiment or run the code? Your report saves the next lab a week.

Sign in to file a reproducibility report.

Related papers

Similarity Map
SelectedSame fieldCross-field
THISDeng '26Hu '25Hu '25Ota '20Zong '22Zhang '24
77% semantic matchBioengineering
Qilin Deng, Guorong Hu, Bradley C. Rauscher+7 more·bioRxiv·Sep 14, 2026