Single-beam driven rotational manipulation for high-resolution 3D cellular morphology reconstruction

Lei Gong; Yijing Wu; Yinmei Li; Yiwei Pan; Ziqiang Wang

REVIEW 3 major objections 4 minor 17 references

Single-beam driven rotational manipulation for high-resolution 3D cellular morphology reconstruction

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read This paper claims that a single focused beam carrying spin angular momentum can simultaneously trap a cell and rotate it about a transverse axis, allowing a coaxially aligned camera to collect the multi-view images from which a visual…

desk verdict A useful application of the group's own single-beam SAM rotation to 3D cell morphology, but the quantitative claims—isotropic resolution and precise volumes—rest on unmeasured rotation poses and a mistaken equation of sampling density with resolution. read the letter →

arxiv 2506.07145 v1 pith:RCF6QQ7W submitted 2025-06-08 physics.optics

Yiwei Pan , Yijing Wu , Ziqiang Wang , Yinmei Li , Lei Gong This is my paper

classification physics.optics

keywords spinangularmomentumopticaltweezerscellrotationmulti-viewimaging3Dreconstructionvisualhullneuralnetworksegmentationisotropicresolution

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper reports a route to full three-dimensional cell morphology that avoids the usual trade-off between flexible rotation and system complexity. The authors use a single focused beam carrying spin angular momentum, the rotational property of light, to both hold a cell in an optical trap and rotate it about a transverse axis, so a camera aligned with the beam captures images from many viewing directions. A neural-network segmentation step turns each image into a clean silhouette, and the visual-hull algorithm combines those silhouettes into a 3D voxel model with equal resolution in every direction. If the approach works as described, any standard holographic optical tweezer setup could perform in situ, label-free 3D reconstruction of microscopic biological specimens.

What carries the argument

The load-bearing mechanism is the spin angular momentum (SAM) vector of a tightly focused optical field. The setup orients a linearly polarized beam at 45° to the polarization axis controlled by a spatial light modulator; the modulator's phase pattern sets the SAM direction of the focused field, and choosing a transverse SAM component makes the trapped cell rotate about a transverse axis, which is what presents many faces to a coaxially aligned camera. On the reconstruction side, the visual hull algorithm is the central object: each binary mask defines a 3D volume, and the intersection of these volumes across viewing angles is the reconstructed cell model. The neural-network preprocessing exists to keep those masks accurate even with slight defocus.

What would settle it

Trap a calibration bead of known diameter, rotate it with the same SAM beam, and track a surface mark in high-speed video to measure the instantaneous rotation axis and angle; then reconstruct the bead with the visual-hull pipeline. If the axis drifts measurably during the recording, or if the reconstructed volume differs from the true volume by more than voxel-quantization error, the fixed-axis assumption behind the method fails.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is an integrated method for isotropic 3D reconstruction of individual cells from images taken while the cell rotates in a single-beam optical trap. A linearly polarized beam shaped by a spatial light modulator acquires a controllable spin angular momentum vector; when tightly focused, the beam's transverse SAM component exerts a torque that spins the cell about a transverse axis while the intensity gradient continues to trap it. A coaxially aligned camera records the resulting multi-view images, neural networks supply binary masks, and the visual hull, the intersection of the volumes swept out by each silhouette, produces a 3D model. The authors demonstrate the workflow on Punica granatum pollen cells and Prunus cerasifera cells, reporting reconstructed volumes of 6021.3 and 1173.1 μm³ with isotropic voxels of 0.1372 μm per side.

Load-bearing premise

The reconstruction assumes each cell is rigid and rotates about a fixed, known transverse axis for every recorded image; the paper does not measure axis stability or the true angle between views, so wobble, precession, or rotation-angle drift would bias the 3D model and volume even if the silhouettes are perfect.

Editorial extensions

If this is right

Standard holographic optical tweezer setups can in principle be upgraded for 3D cell imaging without additional sample stages, flow chambers, or rotating tips.
Full angular coverage eliminates the missing-cone problem, so axial resolution no longer lags behind lateral resolution.
The same single-beam rotation and visual-hull pipeline should transfer to other freestanding microscopic specimens, not just the two cell types demonstrated.
Because the segmentation tolerates slight defocus, the imaging does not demand perfect focal stability during rotation.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Editorial extension: the reported volume values inherit the visual-hull assumption that each cell is rigid and rotates about a fixed axis; a direct measurement of axis wobble would determine how far the quoted volumes can be trusted.
A calibration experiment with a microsphere of known diameter could turn the method into a quantitative 3D morphometer, by comparing reconstructed volume against the true one.
If the SAM-driven rotation remains stable on softer samples, the pipeline might extend to live cells, where rigidity and shape change during rotation would be the main obstacles.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Short version: this is a legitimate application paper, not a physics breakthrough. The rotation mechanism is their own earlier work, but they build a complete workflow—YOLO detection, defocus-robust segmentation, visual hull—and demonstrate it on two real cell types. The qualitative results in Fig. 4 look plausible, and the setup is genuinely simple: one beam, one coaxial camera, no microfluidics. For people working on label-free 3D cell imaging, that combination is worth knowing about.

What's actually new: the integration of the single-beam SAM rotation with a neural-network-based segmentation pipeline and visual hull reconstruction, applied to pollen and plum cells. That's a reasonable and useful contribution, and the masks and projections visibly match the raw images. The use of YOLOv11-style segmentation to handle slight defocus is a sensible engineering choice.

Now the soft spots, which are concentrated in the quantitative claims. The stress-test note lands: the visual hull assumes a known, fixed rotation axis and known per-frame angles, but the paper never measures either. The setup says the cell rotates along (90°, Φ), but no Φ value is reported for any frame. If the cell precesses or the angular increment drifts—which is plausible for SAM-driven rotation of an asymmetric object—the silhouette-based volume is biased even with perfect masks. So the volume estimates 6021.3 and 1173.1 µm³, given to four significant figures, are not established.

Second, the isotropic-resolution claim is stated as a definition: one pixel is 0.1372 µm, so each voxel is that size, therefore that's the resolution. That is sampling density, not resolution. No edge response, no reference object, no measurement of what the smallest resolvable feature is. This needs to be either removed or supported.

Third, there is no ground-truth comparison and no uncertainty quantification. A comparison to a known bead or to a different 3D method would settle whether the visual hull actually recovers morphology accurately.

These are reparable issues. The paper's conceptual framework is sound; the flaws are missing measurements, not wrong physics. Minor clarity issues (e.g., the text referring to Equation (2) when it means Equation (6)) are trivial.

Who gains from this: optical tweezers groups and anyone building label-free multi-view cell imaging systems. I would send it to peer review, but with a strong request for rotation-pose calibration, axis-stability characterization, and a measured resolution test. The claim about isotropic resolution should be revised unless it is actually verified.

Referee Report

3 major / 4 minor

Summary. The paper reports a 3D cell-morphology reconstruction workflow based on single-beam optical tweezers with spin-angular-momentum (SAM) induced rotation. A coaxially aligned camera acquires multi-view images while the cell rotates around a transverse axis. The images are processed with a YOLO-based detection and segmentation pipeline, and the binary masks are fed into a visual hull algorithm to produce a 3D model. The workflow is demonstrated on two cells (Punica granatum pollen and Prunus cerasifera), and cell volumes of 6021.3 and 1173.1 μm3 are computed from the voxel counts.

Significance. The paper tackles an important problem: obtaining isotropic 3D morphology of free-floating cells. Its main strength is the simplicity of the experimental concept—a standard holographic optical tweezer with a coaxially aligned camera—and the integration of a neural-network segmentation stage to deal with defocus artifacts. If the rotation pose is properly calibrated and validated, this could become a practical tool for label-free 3D cell imaging. However, as presented, the central quantitative claims (isotropic resolution, accurate volume) are not supported by the evidence: the rotation geometry is unmeasured, no ground-truth comparison is provided, and the resolution claim conflates pixel sampling with actual resolution.

major comments (3)

[Experiment setup; Algorithm workflow] The visual hull algorithm requires a known 3D pose for every input silhouette, i.e., a fixed rotation axis and a known rotation angle per view. The manuscript states that the trapped cell rotated along (90°, Φ) but does not report any measured value of Φ or the angular increment between frames; the 'rotational alignment' step only orients the masks horizontally or vertically in the image plane. Without pose calibration or an explicit, validated assumption of uniform rotation about a stable axis, the projection geometries feeding the visual hull are undefined, and the reconstructed volumes (6021.3 and 1173.1 μm3) are not established. Please add a calibration measurement (e.g., tracking a reference feature or a known bead) that gives per-frame rotation angles and axis stability, or otherwise demonstrate that the pose is known for every frame.
[Results, paragraph after Fig. 4] The claim that 'each voxel in the reconstructed 3D model corresponds to a volume of 0.1372×0.1372×0.1372 μm3, which defines the isotropic spatial resolution of the system' conflates voxel sampling density with spatial resolution. Sampling density is not resolution; no edge response, known reference object, or comparison with a higher-resolution modality is shown. Furthermore, the resolution of a visual hull depends on the number and angular distribution of views, which are not reported. Please provide an actual resolution test and state the number of views and angular coverage used for each reconstruction.
[Algorithm workflow; Results] No quantitative validation of the reconstruction is presented. The paper reports segmentation masks and reconstructed models, but there is no segmentation accuracy metric (e.g., Dice or IoU), no reprojection error between masks and model projections, and no ground-truth comparison of shape or volume (e.g., against confocal microscopy, SEM, or a calibrated microsphere). In addition, the visual hull is by construction an outer approximation of the object (the intersection of silhouette cones), so voxel-count volumes are systematically biased; this bias and the resulting uncertainty in the reported volumes should be discussed.

minor comments (4)

[Introduction] There is a duplicated phrase: 'rotating cells along arbitrary directions along arbitrary directions' and a missing period after 'These limitations significantly hinder their practical applicability.'
[Experiment setup] The text says the SLM phase is applied 'according to Equation (2)', but the displayed formula is Eq. (6); this should refer to Eq. (4). The relationship between Eq. (4) and Eq. (6) (apparently the specific case Θ=90°) should be stated explicitly.
[Fig. 4] Figure 4 would benefit from clear labels distinguishing the raw image, segmentation mask, and reconstructed-model projection rows, as well as scale bars and axis labels in the cross-sectional panels.
[Methods / Results] Please report the number of multi-view images used for each reconstruction and the total rotation angle covered; this information is needed to assess the visual-hull completeness.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the reconstruction workflow is data-driven and no claimed result reduces to its own inputs.

full rationale

The paper's derivation chain is self-contained in the relevant sense. Equations (1)-(3) are standard textbook expressions for SAM density of a plane wave. Equation (4)/(6) is a reported phase pattern taken from the prior single-beam SAM-rotation method [15], and the experiment uses that method to rotate cells while recording real multi-view images. The reconstruction pipeline then proceeds from measured images: YOLOv8 detection and YOLOv11 segmentation produce binary masks from the recorded frames; rotational alignment and bounding-rectangle centering are image-processing steps; the visual hull algorithm [16] intersects the silhouette volumes; and the final cell volumes are obtained by counting occupied voxels multiplied by the calibrated pixel-to-voxel size. At no point is a target quantity, such as the 3D shape, the volume, or a reported resolution value, inserted as an input or defined in terms of the output. The only overlapping-author citation is [15], which supplies the rotational manipulation method; that citation is not used to justify the reconstruction results, which depend on the measured images and masks. The paper's assertion that the pixel-to-voxel mapping defines 'isotropic spatial resolution' is a resolution/sampling concern rather than a circularity, because the mapping is a calibration statement, not a fitted parameter used to force the reconstruction. No fitted input is renamed as a prediction, and no uniqueness claim is imported to forbid alternatives. Therefore no significant circularity is present.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The central reconstruction depends on two fitted or hand-chosen elements: the rotation axis parameters (Theta, Phi) and the trained YOLO segmentation weights. It also relies on standard physical and algorithmic assumptions about SAM torque and visual hull reconstruction, none of which are invented for this paper.

free parameters (2)

Rotation axis direction (Theta, Phi) = (90 degrees, Phi; exact Phi not reported)
The SLM phase pattern in Eq. (4) is controlled by Theta and Phi to set the SAM direction. In the experiment, Theta=90 degrees is chosen for transverse rotation, but the specific Phi per cell is not reported, and the reconstruction alignment assumes a known rotation axis.
YOLO segmentation model weights = trained on manually annotated in-focus and defocused images
The segmentation masks that drive visual hull reconstruction come from a neural network trained on a manually annotated dataset. The model's accuracy is not quantified, and the weights are not shared, so the reconstruction depends on this fitted model.

assumptions (3)

domain assumption Transfer of SAM from a tightly focused beam exerts torque on a cell, causing rotation about the SAM axis.
Stated in the introduction and relying on standard electrodynamics; the cell is assumed to behave as a birefringent or absorbing particle.
domain assumption The visual hull of silhouettes from multi-view masks is an adequate representation of the cell's 3D morphology.
Standard algorithm from ref [16]; assumes the cell is rigid and silhouettes are exact, but real masks contain errors.
domain assumption Full-angle rotation of the cell about a transverse axis solves the missing cone problem and provides isotropic resolution.
The paper claims isotropic resolution from full angular coverage; this assumes the rotation covers all necessary views and that the visual hull reconstruction preserves that resolution.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Single-beam driven rotational manipulation for high-resolution 3D cellular morphology reconstruction." pith.science (2026). https://pith.science/paper/RCF6QQ7W

@misc{pith2026250607145,
  author       = {Pith},
  title        = {Pith review of: Single-beam driven rotational manipulation for high-resolution 3D cellular morphology reconstruction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RCF6QQ7W}},
  note         = {Machine review of arXiv:2506.07145}
}

read the original abstract

The acquisition of multi-view information of cells is essential for accurate 3D reconstruction of their structures. Rotational manipulation of cells has emerged as an effective technique for obtaining such data. However, most reported methods require a trade-off between manipulation flexibility and system complexity These limitations significantly hinder their practical applicability. Recently, a novel approach has been proposed that enables simultaneous trapping and arbitrary-angle rotation of cells using a single optical beam carrying spin angular momentum (SAM). This method offers improved stability and manipulation flexibility, a simplified experimental setup, and supports coaxial alignment of the imaging and optical paths. In this paper, we employed this method to rotate cells and acquire multi-view images. Furthermore, we present a complete 3D reconstruction workflow, and validate the performance of the proposed method through the reconstruction of Punica granatum pollen cells and Prunus cerasifera cells. Our methods pave the way for 3D reconstruction of microscopic biological specimens, including but not limited to cells.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

17 extracted references · 16 canonical work pages

[1]

J. Lim, K. Lee, K.H. Jin, S. Shin, S. Lee, Y. Park, J.C. Ye, Comparative study of iterative reconstruction algorithms for missing cone problems in optical diffraction tomography, Opt Express, 23 (2015) 16933-16948

work page 2015
[2]

Simon, M

B. Simon, M. Debailleul, M. Houkal, C. Ecoffet, J. Bailleul, J. Lambert, A. Spangenberg, H. Liu, O. Soppera, O. Haeberlé , Tomographic diffractive microscopy with isotropic resolution, Optica, 4 (2017)

work page 2017
[3]

van Rooij, J

J. van Rooij, J. Kalkman, Large-scale high-sensitivity optical diffraction tomography of zebrafish, Biomed Opt Express, 10 (2019) 1782-1793

work page 2019
[4]

K.C. Zhou, R. Qian, S. Degan, S. Farsiu, J.A. Izatt, Optical coherence refraction tomography, Nat Photonics, 13 (2019) 794-802

work page 2019
[5]

K. Kim, J. Yoon, Y. Park, Large-scale optical diffraction tomography for inspection of optical plastic lenses, Optics Letters, 41 (2016) 934-937

work page 2016
[6]

Kvale Lovmo, S

M. Kvale Lovmo, S. Deng, S. Moser, R. Leitgeb, W. Drexler, M. Ritsch-Marte, Ultrasound-induced reorientation for multi-angle optical coherence tomography, Nat Commun, 15 (2024) 2391

work page 2024
[7]

Ahmed, A

D. Ahmed, A. Ozcelik, N. Bojanala, N. Nama, A. Upadhyay, Y. Chen, W. Hanna-Rose, T.J. Huang, Rotational manipulation of single cells and organisms using acoustic waves, Nat Commun, 7 (2016) 11085

work page 2016
[8]

Lee, J.W

J.H. Lee, J.W. Kim, M. Levy, A. Kao, S.H. Noh, D. Bozovic, J. Cheon, Magnetic Nanoparticles for Ultrafast Mechanical Control of Inner Ear Hair Cells, Acs Nano, 8 (2014) 6590-6598

work page 2014

Show all 17 references

[9]

Y. Mao, S. Li, Z. Wang, M. Shao, P. Wang, X. Tan, F. Lu, Y. Wang, X. Wei, Z. Zhong, J. Zhou, Optofluidic-based cell multi-axis controllable rotation and 3D surface imaging, Applied Physics Letters, 123 (2023)

2023
[10]

J. Sun, B. Yang, N. Koukourakis, J. Guck, J.W. Czarske, AI-driven projection tomography with multicore fibre-optic cell rotation, Nat Commun, 15 (2024) 147

2024
[11]

M. Lee, K. Kim, J. Oh, Y. Park, Isotropically resolved label-free tomographic imaging based on tomographic moulds for optical trapping, Light: Science & Applications, 10 (2021) 102

2021
[12]

Merola, L

F. Merola, L. Miccio, P. Memmolo, G. Di Caprio, A. Galli, R. Puglisi, D. Balduzzi, G. Coppola, P. Netti, P. Ferraro, Digital holography as a method for 3D imaging and estimating the biovolume of motile cells, Lab Chip, 13 (2013) 4512-4516

2013
[13]

L. Peng, L. Duan, K. Wang, F. Gao, L. Zhang, G. Wang, Y. Yang, H. Chen, S. Zhang, Transverse photon spin of bulk electromagnetic waves in bianisotropic media, Nature Photonics, 13 (2019) 878-882

2019
[14]

Aiello, P

A. Aiello, P. Banzer, M. Neugebauer, G. Leuchs, From transverse angular momentum to photonic wheels, Nature Photonics, 9 (2015) 789-795

2015
[15]

Y.J. Wu, J.H. Zhuang, P.P. Yu, Y.F. Liu, Z.Q. Wang, Y.M. Li, C.W. Qiu, L. Gong, Time-varying 3D optical torque via a single beam, Nat Commun, 16 (2025) 593. 13

2025
[16]

Laurentini, The Visual Hull Concept for Silhouette-Based Image Understanding, Ieee T Pattern Anal, 16 (1994) 150-162

A. Laurentini, The Visual Hull Concept for Silhouette-Based Image Understanding, Ieee T Pattern Anal, 16 (1994) 150-162

1994
[17]

D. Reis, J. Kupec, J. Hong, A. Daoudi, Real-time flying object detection with YOLOv8, arXiv preprint arXiv:2305.09972, (2023). 14 Authors Contributions L.G. conceived the project. Y.W. designed the system , and Y.P. performed exper iments and analyzed the experimental results....

2023 arXiv

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

J. Lim, K. Lee, K.H. Jin, S. Shin, S. Lee, Y. Park, J.C. Ye, Comparative study of iterative reconstruction algorithms for missing cone problems in optical diffraction tomography, Opt Express, 23 (2015) 16933-16948

work page 2015

[2] [2]

Simon, M

B. Simon, M. Debailleul, M. Houkal, C. Ecoffet, J. Bailleul, J. Lambert, A. Spangenberg, H. Liu, O. Soppera, O. Haeberlé , Tomographic diffractive microscopy with isotropic resolution, Optica, 4 (2017)

work page 2017

[3] [3]

van Rooij, J

J. van Rooij, J. Kalkman, Large-scale high-sensitivity optical diffraction tomography of zebrafish, Biomed Opt Express, 10 (2019) 1782-1793

work page 2019

[4] [4]

K.C. Zhou, R. Qian, S. Degan, S. Farsiu, J.A. Izatt, Optical coherence refraction tomography, Nat Photonics, 13 (2019) 794-802

work page 2019

[5] [5]

K. Kim, J. Yoon, Y. Park, Large-scale optical diffraction tomography for inspection of optical plastic lenses, Optics Letters, 41 (2016) 934-937

work page 2016

[6] [6]

Kvale Lovmo, S

M. Kvale Lovmo, S. Deng, S. Moser, R. Leitgeb, W. Drexler, M. Ritsch-Marte, Ultrasound-induced reorientation for multi-angle optical coherence tomography, Nat Commun, 15 (2024) 2391

work page 2024

[7] [7]

Ahmed, A

D. Ahmed, A. Ozcelik, N. Bojanala, N. Nama, A. Upadhyay, Y. Chen, W. Hanna-Rose, T.J. Huang, Rotational manipulation of single cells and organisms using acoustic waves, Nat Commun, 7 (2016) 11085

work page 2016

[8] [8]

Lee, J.W

J.H. Lee, J.W. Kim, M. Levy, A. Kao, S.H. Noh, D. Bozovic, J. Cheon, Magnetic Nanoparticles for Ultrafast Mechanical Control of Inner Ear Hair Cells, Acs Nano, 8 (2014) 6590-6598

work page 2014