AI-based single-shot structured-light depth reconstruction for real-time laparoscopic surgical guidance

Axel Krieger; Jin U. Kang; Justin Opfermann; Michael Kam; Seonghoon Jang; Shuwen Wei; Wayne Wonseok Rodgers; Xiangyi Le

REVIEW 4 major objections 4 minor 32 references

AI-based single-shot structured-light depth reconstruction for real-time laparoscopic surgical guidance

T0 review · 4 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read An LED-lit passive binary mask plus a VQ-VAE latent-space U-Net recovers metric depth from a single endoscopic frame without projector–camera synchronization, reaching 3.70 mm MAE at 26.0 Hz on a 722-pair phantom dataset.

desk verdict Useful hardware simplification and honest cross-camera supervision, but the headline MAE is not interpretable until the SSLE–Zivid reprojection residual is quantified. read the letter →

arxiv 2608.05109 v1 pith:M6GYZ6YJ submitted 2026-08-05 eess.IV cs.ROphysics.optics

Wayne Wonseok Rodgers , Xiangyi Le , Seonghoon Jang , Shuwen Wei , Justin Opfermann , Michael Kam , Axel Krieger , Jin U. Kang This is my paper

classification eess.IVcs.ROphysics.optics

keywords single-shotfringeprojectionprofilometrybinarystructuredlightendoscopylaparoscopicsurgerydepthestimationroboticVQ-VAElatentpriorvideo-ratereconstruction

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

The paper sets out to show that laparoscopic depth sensing can shed its two heaviest hardware requirements: the digital-micromirror projector and the projector–camera synchronization link. It replaces the projector with a static binary stripe mask illuminated by an LED and trains a VQ-VAE plus U-Net to regress metric depth from one patterned image in a discrete latent space. On 722 paired phantom images, with reference depth supplied by an independent structured-light camera reprojected into the endoscope frame, the method reports 3.70 mm mean absolute error, 0.0326 relative error, and 0.962 $\delta=1.1$ accuracy, running at 26.0 Hz on an NVIDIA A100 GPU. If the claim holds, a compact passive projection module plus a learned latent-space depth prior gives robotic laparoscopy the dense, metric, real-time depth it needs for closed-loop control.

What carries the argument

The mechanism that carries the argument is the pairing of a static binary stripe mask with a VQ-VAE (a vector-quantized variational autoencoder, which compresses an image into discrete tokens from a learned codebook and constrains outputs to the training distribution). The LED-illuminated mask removes temporal synchronization from the optical chain, and the latent-space U-Net performs image-to-depth regression without a segmentation branch. A one-time calibration and z-buffer reprojection of the external camera's depth into the endoscopic frame supplies the pixel-aligned supervision and defines the validity mask used for evaluation.

What would settle it

Acquire depth for rigid targets at known translation-stage positions using a reference measured directly in the endoscope's own image frame, bypassing cross-camera reprojection; if the network's errors in that direct comparison are much smaller than 3.70 mm, the reported MAE was dominated by calibration misalignment, and if they are similar, the network itself is the limiting factor.

Watch

Extended reading notes

Core claim

The central discovery is that a passively patterned LED projection path and a vector-quantized latent-space network together make single-shot structured-light depth estimation synchronization-free and video-rate while dropping the explicit segmentation branch. The VQ-VAE encoder turns the pattern-distorted image into discrete learned codes; a multi-residual U-Net maps those codes to a latent depth representation; and the VQ-VAE decoder renders a dense metric depth map. Supervised on depth from an external structured-light camera reprojected into the endoscope frame, the model reaches 3.70 mm MAE and 0.962 $\delta=1.1$ accuracy on the held-out test set, beats the dual U-Net MaskNet + DepthNet baseline on MAE and relative error, and outperforms off-the-shelf monocular depth models outright. Because every frame carries a valid static pattern, no synchronization is needed, and the network acquires foreground support implicitly rather than through a mask branch.

Load-bearing premise

The result stands on the one-time calibration between the laparoscope and the external structured-light camera being accurate enough that the reprojected reference depth aligns pixel-for-pixel with the endoscopic image; if that alignment is even mildly off, the reported 3.70 mm error becomes a misalignment floor rather than a measure of depth accuracy.

Editorial extensions

If this is right

Every captured frame is a valid pattern frame, so depth can be generated at camera speed with stable latency; the demonstrated 26.0 Hz allows a fixed processing budget for guidance and control loops.
The deployed pipeline no longer needs a DMD projector, projector–camera synchronization, or an explicit mask-prediction branch, simplifying the hardware and the network footprint for a compact laparoscope.
Supervision from an independent depth sensor separates the learning target from the traditional fringe-projection algorithm, but it transfers cross-camera calibration error into both training and evaluation, so absolute errors are only comparable within the same supervision protocol.
Task-specific calibrated networks beat generic monocular depth models by a wide margin on this data, so off-the-shelf monocular depth is not a substitute for structured-light calibration in surgical depth sensing.

Reading between the lines

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

An implication the paper leaves implicit is that the gap between the 2.28 mm MAE under fringe-projection-generated supervision and the 3.70 mm MAE under cross-camera supervision is likely dominated by reprojection misalignment, not by the change in network architecture; a same-optical-path reference would separate these contributions.
A testable extension is to expose the same pipeline to ex vivo tissue: specular highlights, blood, and deformation may push inputs outside the VQ-VAE codebook learned on phantoms, in which case a domain-adaptation or contrast-normalization step would be needed before clinical use.
Because the projected pattern is static and never changes per frame, the system's per-installation calibration cost is the only non-recurring alignment step; the same module could in principle be adapted to any dual-channel endoscope with a free illumination port.
The latent codebook may also enable self-supervised or unpaired adaptation across endoscopes, since the discrete prior encodes the joint statistics of fringe appearance and depth rather than one specific camera.

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

Colleague,

The paper deserves a serious referee, but the headline number should not be trusted as an absolute accuracy until the authors quantify the SSLE–Zivid reprojection residual. The stress-test note lands: Section 2.5 admits any residual (R,t) error creates an irreducible floor in both training and evaluation, and Sections 3.3 and 4 repeat it, but they never measure the floor. If it is 2–3 mm, the 3.70 mm MAE is mostly calibration artifact. That said, the relative comparison across models is on the same ground truth, so the ranking (proposed < MaskNet+DepthNet < monocular) still holds—only the absolute claim is unsubstantiated.

What is genuinely new: the passive LED-illuminated binary mask in the illumination channel of a dual-channel laparoscope, which removes the DMD projector and the projector–camera synchronization link. That is a practical simplification, and the working-distance contrast measurements (C=0.58 at 7 cm, 0.44 at 10 cm) support the feasibility. Using a laterally mounted Zivid camera as independent supervision is the right call; it decouples the learned model from the FPP algorithm that generated ground truth in the authors' prior work. The VQ-VAE latent-space U-Net is a reasonable adaptation of Zuo et al.'s pipeline, and the 26.0 Hz stable frame rate on an A100 is decent for an initial demonstration.

The soft spots are mostly evidential, not architectural. One fixed split, no error bars, no repeated seeds—so we cannot judge variance. The dataset is 722 pairs, which is small by current standards. The valid-mask evaluation is standard, but it hides errors at mask boundaries where the calibration mismatch is worst. The five-example MAE of 3.01 mm in Section 3.3 is clearly labeled as illustrative, but it invites cherry-picking. No public code or data—only 'available on reasonable request'—so the quantitative results are not reproducible. The comparison would be much stronger if they also trained the same network under their previous FPP-derived supervision with identical conditions, which would let them separate the effect of supervision protocol from the network architecture. Without that, the difference from the 2.28 mm in Zuo et al. is confounded.

The authors are honest about these limitations; Sections 2.5, 3.3, and 4 all flag the calibration floor explicitly. That honesty counts, but it does not make the absolute claim quantitative.

Who this is for: people working on surgical depth sensing, structured-light hardware, and learning-based profilometry. It is a useful engineering data point, not a breakthrough. I would send it to peer review, and I would tell the authors that acceptance should hinge on quantifying the reprojection residual (e.g., a checkerboard or static-scene repeatability test) and adding error bars. If they cannot measure the floor, the paper becomes a feasibility study with a credible relative ranking but no certified absolute accuracy. I would not cite it in my own work until the calibration gap is closed.

Referee Report

4 major / 4 minor

Summary. The manuscript proposes a single-shot structured-light laparoscope in which a passive LED-illuminated binary mask replaces an active DMD projector and synchronization link, with depth recovered by a VQ-VAE encoder, a latent-space MRD U-Net, and a VQ-VAE depth decoder. Training and evaluation reference depth is obtained from 722 paired acquisitions of an intestinal phantom, where a laterally mounted Zivid camera is reprojected into the laparoscope frame. On a held-out test subset, the authors report MAE 3.70 mm, AbsRel 0.0326, accuracy at delta=1.1 of 0.962, and a mean inference rate of 26.0 Hz on an NVIDIA A100. The central claim is that this platform enables synchronization-free, video-rate metric depth estimation for robotic laparoscopy without an explicit segmentation stage.

Significance. The hardware simplification—removing the DMD and projector–camera synchronization while retaining a learnable single-shot reconstruction—is a genuine strength, and the use of an independent Zivid reference rather than FPP-derived ground truth decouples the training target from the reconstruction algorithm. The architecture is described in enough detail to be reproduced, and the reporting includes a held-out test set, representative qualitative results, and baseline comparisons. If the calibration floor and the statistical uncertainty of the results are properly quantified, this would be a useful step toward practical single-shot endoscopic depth sensing. However, as presented, the absolute accuracy claim rests on error metrics whose calibration sensitivity is acknowledged but never measured, and the comparative claims are made from a single split without confidence intervals. The paper's significance is therefore conditional on addressing these issues.

major comments (4)

[Sec. 2.5 and Sec. 4, Eqs. (1)-(4)] The manuscript acknowledges in Sec. 2.5 that residual error in the SSLE–Zivid extrinsic calibration (R,t) manifests as a spatial misalignment that introduces an irreducible component of training loss and evaluation error, and Sec. 4 repeats this concern, but no quantitative estimate is provided. Because every reported metric is computed against D_SSLE, which is also the training target, a misalignment of even a few pixels (or 1–2 mm in depth) would contaminate the headline MAE of 3.70 mm and all compared models. In addition, the validity mask M is defined by the reprojection, so its boundary also shifts with calibration error. Please report a calibration-quality measure—for example, RMS checkerboard reprojection error, hold-out alignment error on a planar target, or a sensitivity analysis that perturbs R,t and recomputes Table 1—and, if possible, verify the reference depth against an independently measured known geometry such as a machined plane or cylinder. Without this, the absolute depth-accuracy claim is not interpretable.
[Sec. 3.1, Table 1] All results are reported from a single fixed train/validation/test split of 722 pairs, with no confidence intervals, standard deviations, or repeated-seed variability. The key comparative claim that the proposed model achieves lower MAE than the MaskNet+DepthNet baseline (3.70 vs. 4.30 mm), and the threshold-accuracy trade-off (0.962 vs. 0.965 at delta=1.1), cannot be assessed for statistical significance from the numbers given. Please state the test-set size, and report bootstrap confidence intervals or results over multiple seeds (and ideally multiple splits). This is necessary to support the stated relative advantage and to allow readers to judge whether the observed differences are within noise.
[Sec. 3.1, Table 1] MiDaS and Depth Anything produce scale-ambiguous, non-metric depth; reporting MAE in millimeters and AbsRel without describing any alignment to the Zivid metric scale makes these baseline numbers non-informative. If a least-squares scale/shift alignment was applied on the valid-pixel mask, it should be described; if it was not applied, the comparison should be framed only as relative, not as metric error. Please specify the alignment procedure (or state that none was used) and show the effect of alignment on the reported metrics.
[Sec. 3.2, Fig. 6] The claim of stable video-rate inference at 26.0 Hz is based on a single mean over 301 frames; no standard deviation, percentiles, or histogram statistics are reported, and it is not stated whether the 38.4 ms per-frame figure includes image resizing to 256x256 and preprocessing or only the network forward pass. Since frame-rate stability is part of the central contribution, please provide the full distribution, state the exact timing scope, and ideally report timing on a clinically realistic GPU (or at least acknowledge the A100 limitation in the same paragraph).

minor comments (4)

[Abstract and Sec. 2.4] The abstract states 'acquire 722 reference images' while Sec. 2.4 describes 722 paired SSLE–Zivid acquisitions; please make the wording consistent.
[Eq. (6)] SSIM is applied to masked depth maps D⊗M and D_hat⊗M; please clarify whether SSIM is computed only over valid pixels or over the zero-filled masked image, since zero padding can bias the structural similarity score.
[Sec. 2.5] For reproducibility, please specify the calibration target (checkerboard square size, number of views, illumination conditions) and the MATLAB calibration toolbox version used to obtain (R,t).
[Fig. 6] The inference-frequency histogram lacks axis labels and units; adding them would make the plot self-contained.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the supervised depth claim is evaluated against an independent Zivid reference on a held-out test split, and the acknowledged calibration floor is an accuracy limitation, not a circular derivation.

full rationale

The paper's central quantitative claims (MAE 3.70 mm, AbsRel 0.0326, delta=1.1 accuracy 0.962) are computed on a held-out test set against reference depth maps from an independent Zivid structured-light camera reprojected into the SSLE frame through a one-time extrinsic calibration. The training target and evaluation metric use the same reprojected Zivid depth, but that is standard supervised learning rather than circularity: Zivid is an external sensor, the predictions are not used to define the reference, and no equation defines the reported result in terms of a fitted parameter that was itself obtained from the same predictions. The VQ-VAE codebooks and U-Net are trained under this supervision and evaluated on unseen samples, so the result is not true by construction. The paper's self-citations to prior group work (Zuo et al. 2025; Rodgers et al. 2026) motivate the hardware direction and the MaskNet + DepthNet baseline, but the novel claims rest on the independent Zivid-referenced protocol, not on those citations. The explicitly acknowledged SSLE-Zivid reprojection misalignment sets a floor on both training loss and evaluation error; this is a real measurement-validity and calibration-accuracy concern, but it is not a circularity because external ground truth remains the source of supervision. No load-bearing step reduces, by the paper's own equations or by self-citation, to its own inputs. Thus no significant circularity is present.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

No new physical entities, particles, forces, or dimensions are introduced. The VQ-VAE latent space, the validity mask, and the binary mask are constructs or methods from prior literature or calibration, not new postulated entities. The paper's contribution is a combination of known hardware and network components.

free parameters (5)

VQ-VAE codebook sizes = K_img=512, K_depth=1024
Chosen by hand without ablation; the latent capacity directly controls how much image and depth structure the prior can represent (Section 2.6).
Loss weighting alpha and beta schedule = alpha not specified; beta_start and beta_end not specified
Equations (5) and (6) rely on hand-set weights; no sensitivity analysis is reported.
Depth normalization range = [0, 250] mm
The decoder rescales tanh output to this range and the SSIM loss normalizes by the global dataset range; depth outside the range is clipped, so the choice affects all metrics (Sections 2.6 and 2.7).
Training hyperparameters = U-Net LR 1e-3, Tmax 160, min 1e-6, batch 4, epochs 161; VQ-VAE LR 5e-5, batch 4, grad accum 4
Standard but hand-picked; no ablations, so sensitivity is unknown (Section 2.7).
Binary mask duty cycle and pattern frequency = Target 1:1 duty cycle; pattern frequency not quantified
Hardware design choice that determines pattern visibility and depth sensitivity; only contrast values at 7 and 10 cm are reported (Section 2.3).

assumptions (6)

standard math Pinhole camera model and rigid-body transform equations (1)-(3) accurately describe both cameras.
Used for Zivid-to-SSLE reprojection; standard but assumes no significant lens distortion or unmodeled optical effects (Section 2.5).
domain assumption The laterally mounted Zivid depth maps are sufficiently accurate and dense to serve as reference depth.
No independent characterization of Zivid error on the phantom is provided; the paper acknowledges that reprojection mismatch sets a floor on metrics (Sections 2.5 and 4).
domain assumption The SSLE-Zivid extrinsic calibration (R,t) is accurate enough for pixel-level alignment.
Authors state that residual calibration error creates an irreducible component of training and evaluation error (Sections 2.5, 3.3, and 4).
domain assumption The phantom surface reflects the binary pattern sufficiently for depth cues, and phantom behavior transfers to tissue.
The paper explicitly notes that ex vivo and in vivo validation remains necessary (Section 4).
ad hoc to paper The VQ-VAE discrete latent space preserves depth-relevant information after quantization.
Central architectural bet; no analysis of information loss, codebook usage, or reconstruction fidelity is provided (Section 2.6).
domain assumption The validity mask M defines an unbiased evaluation region.
Metrics are computed only where Zivid reprojects valid depth; if masked-out pixels are systematically harder, reported MAE may be optimistic (Section 2.5, Equation 9).

how reviews work

0 comments

Cite this review

Pith. "Pith review of AI-based single-shot structured-light depth reconstruction for real-time laparoscopic surgical guidance." pith.science (2026). https://pith.science/paper/M6GYZ6YJ

@misc{pith2026260805109,
  author       = {Pith},
  title        = {Pith review of: AI-based single-shot structured-light depth reconstruction for real-time laparoscopic surgical guidance},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/M6GYZ6YJ}},
  note         = {Machine review of arXiv:2608.05109}
}

read the original abstract

Significance. Accurate intraoperative depth perception is important for autonomous and semi-autonomous robotic laparoscopic surgery. Conventional fringe projection profilometry can achieve millimeter-scale accuracy but often requires multi-shot acquisition, digital-micromirror-device projection, and projector-camera synchronization, complicating integration into compact laparoscopic systems. Aim. To develop a synchronization-free, single-shot depth-sensing platform using a passive LED-illuminated binary mask and a VQ-VAE prior with a custom U-Net depth head. Approach. A compact projection module was coupled to one channel of a dual-channel laparoscope, while the second channel imaged the fringe-illuminated target. A Zivid 3D camera acquired reference depth for 722 paired phantom images. Zivid depth maps were reprojected into the SSLE image frame for supervised training and evaluation. The VQ-VAE encoded each input into a discrete latent representation, and a latent-space U-Net predicted depth without a separate mask-prediction branch. Results. Using a fixed train/validation/test split, the proposed model achieved an MAE of 3.70 mm, AbsRel of 0.0326, delta=1.1 accuracy of 0.962, and delta=1.1^2 accuracy of 0.970. It achieved lower MAE than the dual U-Net MaskNet + DepthNet baseline and outperformed off-the-shelf monocular depth models in MAE, AbsRel, and threshold accuracy. The pipeline operated at 26.0 Hz over 301 consecutive frames on an NVIDIA A100 GPU. Conclusions. The LED-illuminated binary-pattern platform with latent-space depth reconstruction enables synchronization-free, video-rate endoscopic depth estimation. Results demonstrate Zivid-referenced phantom reconstruction without an explicit segmentation stage, while emphasizing the importance of dataset size and SSLE-Zivid calibration accuracy.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

32 extracted references · 30 canonical work pages

[1]

and Opfermann, J

Saeidi, H. and Opfermann, J. D. and Kam, M. and Wei, S. and Leonard, S. and Hsieh, M. H. and Kang, J. U. and Krieger, A. , title =. Sci. Robot. , volume =. 2022 , doi =

work page 2022
[2]

and Deguet, Anton and Kang, Jin U

Song, Hyunwoo and Moradi, Hamid and Jiang, Baichuan and Xu, Keshuai and Wu, Yixuan and Taylor, Russell H. and Deguet, Anton and Kang, Jin U. and Salcudean, Septimiu E. and Boctor, Emad M. , title =. IEEE Robot. Autom. Lett. , volume =. 2023 , doi =

work page 2023
[3]

and Sunmola, Idris and Hsieh, Michael H

Wang, Yaning and Wei, Shuwen and Zuo, Ruizhi and Kam, Michael Clarification and Opfermann, Justin D. and Sunmola, Idris and Hsieh, Michael H. and Krieger, Axel and Kang, Jin U. , title =. Biomed. Opt. Express , volume =. 2024 , doi =

work page 2024
[4]

Zhang, Song , title =. Opt. Lasers Eng. , volume =. 2010 , doi =

work page 2010
[5]

Geng, Jason , title =. Adv. Opt. Photonics , volume =. 2011 , doi =

work page 2011
[6]

Le, Hanh N. D. and Nguyen, Hieu and Wang, Zhaoyang and Opfermann, Justin D. and Leonard, Simon and Krieger, Axel and Kang, Jin U. , title =. J. Biomed. Opt. , volume =. 2018 , doi =

work page 2018
[7]

and Saeidi, Hamed and Hsieh, Michael H

Wei, Shuwen and Kam, Michael and Wang, Yaning and Opfermann, Justin D. and Saeidi, Hamed and Hsieh, Michael H. and Krieger, Axel and Kang, Jin U. , title =. J. Opt. Soc. Am. A , volume =. 2022 , doi =

work page 2022
[8]

Zuo, Chao and Feng, Shijie and Huang, Lei and Tao, Tianyang and Yin, Wei and Chen, Qian , title =. Opt. Lasers Eng. , volume =. 2018 , doi =

work page 2018

Show all 32 references

[9]

Photonics , volume =

Nguyen, Hieu and Wang, Zhaoyang , title =. Photonics , volume =. 2021 , doi =

2021
[10]

Wang, Fei and Wang, Chao and Guan, Qun , title =. Opt. Express , volume =. 2021 , doi =

2021
[11]

and Wei, S

Zuo, R. and Wei, S. and Wang, Y. and Huang, R. and Rodgers, W. W. and Yu, J. and Hsieh, M. H. and Krieger, A. and Kang, J. U. , title =. J. Biomed. Opt. , volume =. 2025 , doi =

2025
[12]

Rodgers, W. W. and Zuo, R. and Yu, J. and Jang, S. and Wei, S. and Opfermann, J. and Krieger, A. and Kang, J. U. , title =. Advanced Biomedical and Clinical Diagnostic and Surgical Guidance Systems XXIV , editor =. 2026 , doi =

2026
[13]

Advances in Neural Information Processing Systems (NeurIPS) , volume =

van den Oord, Aaron and Vinyals, Oriol and Kavukcuoglu, Koray , title =. Advances in Neural Information Processing Systems (NeurIPS) , volume =. 2017 , url =

2017
[14]

Medical Image Computing and Computer-Assisted Intervention -- MICCAI 2015 , series =

Ronneberger, Olaf and Fischer, Philipp and Brox, Thomas , title =. Medical Image Computing and Computer-Assisted Intervention -- MICCAI 2015 , series =. 2015 , doi =

2015
[15]

Eigen, David and Puhrsch, Christian and Fergus, Rob , title =. Adv. Neural Inf. Process. Syst. , volume =. 2014 , url =

2014
[16]

and Bovik, A

Wang, Z. and Bovik, A. C. and Sheikh, H. R. and Simoncelli, E. P. , title =. IEEE Trans. Image Process. , volume =. 2004 , doi =

2004
[17]

Advances in Neural Information Processing Systems , volume =

Paszke, Adam and Gross, Sam and Massa, Francisco and Lerer, Adam and Bradbury, James and Chanan, Gregory and Killeen, Trevor and Lin, Zeming and Gimelshein, Natalia and Antiga, Luca and Desmaison, Alban and K. Advances in Neural Information Processing Systems , volume =. 2019 , doi =

2019
[18]

and Ba, Jimmy , title =

Kingma, Diederik P. and Ba, Jimmy , title =. International Conference on Learning Representations (ICLR) , year =
[19]

and Saeidi, H

Kam, M. and Saeidi, H. and Wei, S. and Opfermann, J. D. and L. Semi-autonomous robotic anastomoses of vaginal cuffs using marker enhanced. Medical Image Computing and Computer-Assisted Intervention -- MICCAI 2019 , series =. 2019 , doi =

2019
[20]

and Wang, Zhaoyang and Kieu, Hien and Le, Minh , title =

Nguyen, Hieu and Nguyen, Dung A. and Wang, Zhaoyang and Kieu, Hien and Le, Minh , title =. Appl. Opt. , volume =. 2015 , doi =

2015
[21]

Du, Hua and Wang, Zhenghe , title =. Opt. Lett. , volume =. 2007 , doi =

2007
[22]

and Wang, Zhaoyang and Pan, Bing and Pan, Tianyan , title =

Vo, Minh N. and Wang, Zhaoyang and Pan, Bing and Pan, Tianyan , title =. Opt. Express , volume =. 2012 , doi =

2012
[23]

and Wang, Zhaoyang and Luu, Long and Ma, Jun , title =

Vo, Minh N. and Wang, Zhaoyang and Luu, Long and Ma, Jun , title =. Opt. Eng. , volume =. 2011 , doi =

2011
[24]

Clancy, N. T. and Stoyanov, D. and Maier-Hein, L. and Groch, A. and Yang, G.-Z. and Elson, D. S. , title =. Biomed. Opt. Express , volume =. 2011 , doi =

2011
[25]

Saldner, H. O. and Huntley, J. M. , title =. Opt. Eng. , volume =. 1997 , doi =

1997
[26]

Ackerman, J. D. and Keller, K. and Fuchs, H. , title =. Proc. SPIE , volume =. 2002 , doi =

2002
[27]

and Kwartowitz, David M

Ronaghi, Zahra and Duffy, Edward B. and Kwartowitz, David M. , title =. Journal of Medical Imaging , volume =. 2015 , doi =

2015
[28]

Sensors , volume =

Khoshelham, Kourosh and Elberink, Sander Oude , title =. Sensors , volume =. 2012 , doi =

2012
[29]

Assessment and calibration of a

Lachat, Elise and Macher, H. Assessment and calibration of a. Remote Sens. , volume =. 2015 , doi =

2015
[30]

Maier-Hein and A

L. Maier-Hein and A. Groch and A. Bartoli and S. Bodenstedt and G. Boissonnat and P.-L. Chang and N. T. Clancy and D. S. Elson and S. Haase and E. Heim and J. Hornegger and P. Jannin and H. Kenngott and T. Kilgus and B. M. Comparative validation of single-shot optical techniqu...

2014
[31]

Feng and Q

S. Feng and Q. Chen and G. Gu and T. Tao and L. Zhang and Y. Hu and W. Yin and C. Zuo , title =. Adv. Photonics , volume =. 2019 , doi =

2019
[32]

Liu and A

X. Liu and A. Sinha and M. Ishii and G. D. Hager and A. Reiter and R. H. Taylor and M. Unberath , title =. IEEE Trans. Med. Imaging , volume =. 2020 , doi =

2020

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

and Opfermann, J

Saeidi, H. and Opfermann, J. D. and Kam, M. and Wei, S. and Leonard, S. and Hsieh, M. H. and Kang, J. U. and Krieger, A. , title =. Sci. Robot. , volume =. 2022 , doi =

work page 2022

[2] [2]

and Deguet, Anton and Kang, Jin U

Song, Hyunwoo and Moradi, Hamid and Jiang, Baichuan and Xu, Keshuai and Wu, Yixuan and Taylor, Russell H. and Deguet, Anton and Kang, Jin U. and Salcudean, Septimiu E. and Boctor, Emad M. , title =. IEEE Robot. Autom. Lett. , volume =. 2023 , doi =

work page 2023

[3] [3]

and Sunmola, Idris and Hsieh, Michael H

Wang, Yaning and Wei, Shuwen and Zuo, Ruizhi and Kam, Michael Clarification and Opfermann, Justin D. and Sunmola, Idris and Hsieh, Michael H. and Krieger, Axel and Kang, Jin U. , title =. Biomed. Opt. Express , volume =. 2024 , doi =

work page 2024

[4] [4]

Zhang, Song , title =. Opt. Lasers Eng. , volume =. 2010 , doi =

work page 2010

[5] [5]

Geng, Jason , title =. Adv. Opt. Photonics , volume =. 2011 , doi =

work page 2011

[6] [6]

Le, Hanh N. D. and Nguyen, Hieu and Wang, Zhaoyang and Opfermann, Justin D. and Leonard, Simon and Krieger, Axel and Kang, Jin U. , title =. J. Biomed. Opt. , volume =. 2018 , doi =

work page 2018

[7] [7]

and Saeidi, Hamed and Hsieh, Michael H

Wei, Shuwen and Kam, Michael and Wang, Yaning and Opfermann, Justin D. and Saeidi, Hamed and Hsieh, Michael H. and Krieger, Axel and Kang, Jin U. , title =. J. Opt. Soc. Am. A , volume =. 2022 , doi =

work page 2022

[8] [8]

Zuo, Chao and Feng, Shijie and Huang, Lei and Tao, Tianyang and Yin, Wei and Chen, Qian , title =. Opt. Lasers Eng. , volume =. 2018 , doi =

work page 2018