Single-shot volumetric fluorescence imaging with neural fields

Oumeng Zhang1,*, Haowen Zhou1,*, Brandon Y. Feng2,*,
Elin M. Larsson1, Reinaldo E. Alcalde1, Siyuan Yin1, Catherine Deng1, Changhuei Yang1
1California Institute of Technology, 2Massachusetts Institute of Technology, *Equal contribution
arXiv, 2024
arXiv Code to be released

System overview

FPM-INR

Abstract

Single-shot volumetric fluorescence (SVF) imaging captures biological processes with high temporal resolution and a large field of view, unlike traditional methods requiring multiple axial plane scans. Existing SVF methods often face limitations due to large, complex point spread functions (PSFs), affecting signal-to-noise ratio, resolution, and field of view. The paper introduces the QuadraPol PSF combined with neural fields, using a compact custom polarizer and a polarization camera to detect fluorescence and encode the 3D scene within a compact PSF without depth ambiguity. This approach, coupled with a novel reconstruction algorithm, significantly reduces acquisition time by approximately 20 times and captures a 100 mm³ volume in one shot, demonstrated through imaging bacterial colonies and plant root morphology.

Bacteria Colony on Sand Surfaces

All-in-focus image

sand_all_in_focus

Focal stack

sand_stack

Plant root volumetric imaging


Image volume with color-coded depth

fig_root

Zoom-in slices

Comparison of neural fields and deconvolution for zoom-in regions (xy, xz planes) of the plant root. Depth infomation is encoded in colors.


Deconvolution
Neural fields
Deconvolution
Neural fields


Focal stack for plant root

Loading...



BibTeX

@misc{zhang2024singleshot,
      title={Single-shot volumetric fluorescence imaging with neural fields},
      author={Oumeng Zhang and Haowen Zhou and Brandon Y. Feng and Elin M. Larsson and Reinaldo E. Alcalde and Siyuan Yin and Catherine Deng and Changhuei Yang},
      year={2024},
      eprint={2405.10463},
      archivePrefix={arXiv},
      primaryClass={physics.optics}
  }