pith. sign in

arxiv: 1907.06727 · v1 · pith:GKRDT63Lnew · submitted 2019-07-15 · 📡 eess.IV · cs.CV· cs.LG· physics.optics

Deep learning-based color holographic microscopy

Tairan Liu , Zhensong Wei , Yair Rivenson , Kevin De Haan , Yibo Zhang , Yichen Wu , Aydogan Ozcan This is my paper

Pith reviewed 2026-05-24 21:01 UTC · model grok-4.3

classification 📡 eess.IV cs.CVcs.LGphysics.optics
keywords deep learningGANholographic microscopycolor reconstructionsingle hologramhistopathologytissue imagingmulti-wavelength illumination
0
0 comments X

The pith

A GAN reconstructs high-fidelity color images from one three-wavelength hologram.

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

The paper shows that a generative adversarial network can take a single hologram recorded under simultaneous red, green, and blue illumination and output an accurate color image of stained tissue. It learns to suppress the phase-missing artifacts that normally appear in holographic reconstruction and to apply the correct color mapping without needing separate captures for each wavelength. A reader would care because standard color holographic imaging requires multiple holograms or sequential illuminations, which lowers speed; collapsing the process to one exposure could raise the practical throughput of coherent microscopes used on histological samples.

Core claim

A generative adversarial network trained on hologram-color image pairs learns to remove missing-phase artifacts and to perform an accurate color transformation, enabling high-fidelity color reconstruction from a single hologram of a sample illuminated simultaneously at three wavelengths; the approach is shown on lung and prostate tissue sections carrying different histological stains.

What carries the argument

A generative adversarial network that maps a single three-wavelength hologram to a full-color image while correcting phase artifacts and color balance.

If this is right

  • Only a single hologram exposure is required for accurate color imaging of stained tissue.
  • Throughput of coherent microscopy systems increases because sequential wavelength captures are eliminated.
  • The method applies to point-of-care histopathology on differently stained sections.
  • Color transformation and artifact removal are learned jointly rather than handled by separate post-processing steps.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same network architecture could be retrained on other coherent imaging modalities that combine multiple wavelengths in one shot.
  • Hardware designs that illuminate with three wavelengths simultaneously become more attractive once reconstruction is reliable.
  • Extension to dynamic or live specimens would require only that new training pairs be collected under the same illumination geometry.

Load-bearing premise

The trained network generalizes to new tissue samples and illumination conditions without adding artifacts or color errors.

What would settle it

Reconstruction of a fresh set of lung or prostate sections never seen during training that shows visible color shifts or phase-induced artifacts would falsify the claim.

Figures

Figures reproduced from arXiv: 1907.06727 by Aydogan Ozcan, Kevin De Haan, Tairan Liu, Yair Rivenson, Yibo Zhang, Yichen Wu, Zhensong Wei.

Figure 1
Figure 1. Figure 1: Comparison between the traditional hyperspectral imaging and the proposed neural network-based approaches for the recon￾struction of accurate color images. NH is the number of sample-to-sensor heights required for performing phase recovery, NW is the number of illumination wavelengths, NM is the number of measurements for each illumination condition (multiplexed or sequential), and L is the number of later… view at source ↗
Figure 4
Figure 4. Figure 4: Deep learning-based accurate color imaging of a lung tissue slide stained with Masson’s trichrome for a multiplexed illumination at 450 nm, 540 nm, and 590 nm, using a lens-free holographic on-chip microscope. (a): Large field of view of the network output image. (b): Zoomed-in comparison of the network input, the network output, and the ground truth target at region of interest (ROI) 1 and 2 [PITH_FULL_I… view at source ↗
Figure 5
Figure 5. Figure 5: Deep learning-based accurate color imaging of a prostate tissue slide stained with H&E for a multiplexed illumination at 450 nm, 540 nm, and 590 nm, using a lens-free holographic on-chip microscope. (a): Large field of view of the network output image. (b) Zoomed-in comparison of the network input, the network output, and the ground truth target at ROI 1 and 2 [PITH_FULL_IMAGE:figures/full_fig_p020_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Stitched image of the deep neural network output for a lung tissue section stained with H&E, which corresponds to the sensor’s field-of-view [PITH_FULL_IMAGE:figures/full_fig_p021_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Visual comparison between the deep neural network-based approach and the multiheight phase recovery with spectral estimation approach for a lung tissue sample stained with Masson’s trichrome. (a-h): Reconstruction results of spectral estimation approach using different number of heights and different illumination conditions. (i): Network output. (j): Ground truth target obtained using the hyper￾spectral im… view at source ↗
Figure 8
Figure 8. Figure 8: Visual comparison between the deep neural network-based approach and the multiheight phase recovery with the spectral esti￾mation approach for a prostate tissue sample stained with H&E. (a-h): Reconstruction results of spectral estimation approach using differ￾ent number of heights and different illumination conditions. (i): Network output. (j): Ground truth target obtained using the hyperspectral imaging … view at source ↗
read the original abstract

We report a framework based on a generative adversarial network (GAN) that performs high-fidelity color image reconstruction using a single hologram of a sample that is illuminated simultaneously by light at three different wavelengths. The trained network learns to eliminate missing-phase-related artifacts, and generates an accurate color transformation for the reconstructed image. Our framework is experimentally demonstrated using lung and prostate tissue sections that are labeled with different histological stains. This framework is envisaged to be applicable to point-of-care histopathology, and presents a significant improvement in the throughput of coherent microscopy systems given that only a single hologram of the specimen is required for accurate color imaging.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The paper claims a GAN-based framework for high-fidelity color holographic image reconstruction from a single multi-wavelength hologram (three simultaneous illuminations), where the network learns to remove missing-phase artifacts and apply accurate color transformation. The approach is experimentally demonstrated on lung and prostate tissue sections labeled with different histological stains, with the goal of enabling higher-throughput coherent microscopy for point-of-care histopathology.

Significance. If the generalization and artifact-free performance hold, the result would be significant for computational imaging by reducing the number of required holograms from multiple to one while producing color output, directly addressing throughput limitations in holographic microscopy. The experimental use of real stained histological samples provides a concrete demonstration of the method on relevant data.

major comments (2)
  1. [Abstract / experimental demonstration] Abstract and experimental section: the central claim of applicability to point-of-care histopathology for arbitrary samples rests on generalization of the GAN mapping, yet the reported experiments are restricted to lung and prostate sections with specific stains; no cross-validation, hold-out tissue types, or stain-variation tests are described to rule out memorization of training-distribution color statistics or hologram patterns.
  2. [Abstract] Abstract: the soundness of the high-fidelity claim cannot be assessed because no quantitative validation metrics (e.g., PSNR, SSIM, color error), error analysis, or training/validation split details are provided, leaving the weakest assumption (generalization without new artifacts on unseen samples) untested.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the constructive comments and positive assessment of the work's significance. We provide point-by-point responses to the major comments below and have revised the manuscript to address concerns where feasible by adding clarifications and quantitative details.

read point-by-point responses

  1. Referee: [Abstract / experimental demonstration] Abstract and experimental section: the central claim of applicability to point-of-care histopathology for arbitrary samples rests on generalization of the GAN mapping, yet the reported experiments are restricted to lung and prostate sections with specific stains; no cross-validation, hold-out tissue types, or stain-variation tests are described to rule out memorization of training-distribution color statistics or hologram patterns.

    Authors: The experiments are restricted to lung and prostate sections with different histological stains, which provide variation in color mapping. We have revised the abstract and added a discussion section clarifying the scope of the demonstrations and explicitly noting that the applicability to arbitrary samples is envisaged rather than fully validated. The use of multiple stains helps mitigate concerns of pure memorization, but we acknowledge the absence of hold-out tissue-type testing. revision: partial

  2. Referee: [Abstract] Abstract: the soundness of the high-fidelity claim cannot be assessed because no quantitative validation metrics (e.g., PSNR, SSIM, color error), error analysis, or training/validation split details are provided, leaving the weakest assumption (generalization without new artifacts on unseen samples) untested.

    Authors: We agree that quantitative metrics strengthen the claims. The revised manuscript now includes PSNR, SSIM, and color error metrics, along with error analysis and explicit details on the training/validation splits (e.g., data partitioning from the acquired holograms). These were computed on the existing experimental data to support the high-fidelity assessment. revision: yes

standing simulated objections not resolved
  • Additional cross-validation or tests on hold-out tissue types and further stain variations, as this would require new experimental acquisitions beyond the current dataset.

Circularity Check

0 steps flagged

No circularity: standard supervised GAN training against external ground truth

full rationale

The paper presents a GAN framework for reconstructing color images from single multi-wavelength holograms. The approach relies on supervised training where the network learns a mapping from hologram inputs to color outputs using paired experimental data from lung and prostate sections. No equations, parameters, or claims reduce the output to a fitted input by construction, nor do any load-bearing steps depend on self-citations or imported uniqueness theorems. The central result is an empirical demonstration of the trained model's performance on the reported samples, which is self-contained and falsifiable against held-out measurements.

Axiom & Free-Parameter Ledger

1 free parameters · 0 axioms · 0 invented entities

The central claim rests on a trained neural network whose weights are learned from data; no additional free parameters, axioms, or invented entities are introduced beyond standard GAN training assumptions.

free parameters (1)
  • GAN network weights and hyperparameters
    Learned during training on the specific tissue dataset; central to the reconstruction performance.

pith-pipeline@v0.9.0 · 5650 in / 984 out tokens · 29767 ms · 2026-05-24T21:01:54.961198+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

42 extracted references · 42 canonical work pages · 3 internal anchors

  1. [1]

    Color management in digital pathology,

    W. C. Revie, M. Shires, P. Ja ckson, D. Brettle, R. Cochrane, and D. Treanor, "Color management in digital pathology," Analytical Cellular Pathology 2014, (2014)

    work page 2014

  2. [2]

    Sixty - five thousand shades of gra y: importance of color in surgical pathology diagnoses,

    W. S. Campbell, G. A. Talmon, K. W. Foster, S. M. Lele, J. A. Kozel, and W. W. West, "Sixty - five thousand shades of gra y: importance of color in surgical pathology diagnoses," Human p a- thology 46, 1945–1950 (2015)

    work page 1945

  3. [3]

    Color standardization in whole slide imaging using a color calibration slide,

    P. A. Bautista, N. Hashimoto, and Y. Yagi, "Color standardization in whole slide imaging using a color calibration slide," Journal of pathology informatics 5, (2014)

    work page 2014

  4. [4]

    Color accuracy and reproducibility in whole slide imaging scanners,

    P. Shrestha and B. Hulsken, "Color accuracy and reproducibility in whole slide imaging scanners," Journal of Medical Imaging 1, 027501 (2014)

    work page 2014

  5. [5]

    Wavelength selection for true -color holography,

    M. S. Peercy and L. Hesselink, "Wavelength selection for true -color holography," Applied optics 33, 6811–6817 (1994)

    work page 1994

  6. [6]

    A review of rgb color spaces... from xyy to r’g’b’,

    D. Pascale, "A review of rgb color spaces... from xyy to r’g’b’," Babel Color 18, 136–152 (2003)

    work page 2003

  7. [7]

    A c- curate color imaging of pathology slides using holography and absorbance spectrum estimation of histochemical stains,

    Y. Zhang, T. Liu, Y. Huang, D. Teng, Y. Bian, Y. Wu, Y. Rivenson, A. Feizi, and A. Ozcan, "A c- curate color imaging of pathology slides using holography and absorbance spectrum estimation of histochemical stains," Journal of Biophotonics e201800335 (2018)

    work page 2018

  8. [8]

    Hyperspectral digital holography of microob- jects,

    S. G. Kalenkov, G. S. Kalenkov, and A. E. Shtanko, "Hyperspectral digital holography of microob- jects," in Practical Holography XXIX: Materials and Applications (International Society for Optics and Photonics, 2015), Vol. 9386, p. 938604

    work page 2015

  9. [9]

    Digital Holography Using Spe ctral Estimation Technique,

    P. Xia, Y. Ito, Y. Shimozato, T. Tahara, T. Kakue, Y. Awatsuji, K. Nishio, S. Ura, T. Kubota, and O. Matoba, "Digital Holography Using Spe ctral Estimation Technique," J. Display Technol., JDT 10, 235–242 (2014)

    work page 2014

  10. [10]

    Imaging without lenses: achievements and remaining challenges of wide -field on- chip microscopy,

    A. Greenbaum, W. Luo, T.-W. Su, Z. Göröcs, L. Xue, S. O. Isikman, A. F. Coskun, O. Mudanyali, and A. Ozcan, "Imaging without lenses: achievements and remaining challenges of wide -field on- chip microscopy," Nat. Methods 9, 889–895 (2012)

    work page 2012

  11. [11]

    Phase retrieval from series of images obtained by defocus variation,

    L. J. Allen and M. P. Oxley, "Phase retrieval from series of images obtained by defocus variation," Optics Communications 199, 65–75 (2001)

    work page 2001

  12. [12]

    Wide - field computational imaging of pathology slides using lens -free on -chip microscopy,

    A. Greenbaum, Y. Zhang, A. Feizi, P. -L. Chung, W. Luo, S. R. Kandukuri, and A. Ozcan, "Wide - field computational imaging of pathology slides using lens -free on -chip microscopy," Science Translational Medicine 6, 267ra175-267ra175 (2014)

    work page 2014

  13. [13]

    Maskless imaging of dense samples using pixel super -resolution based multi-height lensfree on-chip microscopy,

    A. Greenbaum and A. Ozcan, "Maskless imaging of dense samples using pixel super -resolution based multi-height lensfree on-chip microscopy," Opt. Express, OE 20, 3129–3143 (2012)

    work page 2012

  14. [14]

    Field -portable wide-field microscopy of dense samples using multi -height pixel super -resolution based lensfree imaging,

    A. Greenbaum, U. Sikora, and A. Ozcan, "Field -portable wide-field microscopy of dense samples using multi -height pixel super -resolution based lensfree imaging," Lab Chip 12, 1242 –1245 (2012)

    work page 2012

  15. [15]

    Sparsity -based multi-height phase recovery in holographic microscopy,

    Y. Rivenson, Y. Wu, H. Wang, Y. Zhang, A. Feizi, and A. Ozcan, "Sparsity -based multi-height phase recovery in holographic microscopy," Scientific Reports 6, 37862 (2016)

    work page 2016

  16. [16]

    Synthetic aperture -based on-chip microscopy,

    W. Luo, A. Greenbaum, Y. Zhang, and A. Ozcan, "Synthetic aperture -based on-chip microscopy," Light: Science & Applications 4, e261 (2015)

    work page 2015

  17. [17]

    Pixel super -resolution using wavelength scanning,

    W. Luo, Y. Zhang, A. Feizi, Z. Göröcs, and A. Ozcan, "Pixel super -resolution using wavelength scanning," Light Sci Appl. 5, e16060 (2016)

    work page 2016

  18. [18]

    Lensfree on -chip microscopy over a wide field-of-view using pixel super-resolution,

    W. Bishara, T. -W. Su, A. F. Coskun, and A. Ozcan, "Lensfree on -chip microscopy over a wide field-of-view using pixel super-resolution," Optics Express 18, 11181 (2010)

    work page 2010

  19. [19]

    Hol o- graphic pixel super -resolution in portable lensless on -chip microscopy using a fiber -optic array,

    W. Bishara, U. Sikora, O. Mudanyali, T. -W. Su, O. Yaglidere, S. Luckhart, and A. Ozcan, "Hol o- graphic pixel super -resolution in portable lensless on -chip microscopy using a fiber -optic array," Lab on a Chip 11, 1276 (2011)

    work page 2011

  20. [20]

    Maskl ess imaging of dense samples using pixel super -resolution based multi-height lensfree on-chip microscopy,

    A. Greenbaum and A. Ozcan, "Maskl ess imaging of dense samples using pixel super -resolution based multi-height lensfree on-chip microscopy," Optics Express 20, 3129 (2012)

    work page 2012

  21. [21]

    Field -portable wide-field microscopy of dense samples using multi-height pixel super-resolution based lensfree imaging,

    A. Greenbaum, U. Sikora, and A. Ozcan, "Field -portable wide-field microscopy of dense samples using multi-height pixel super-resolution based lensfree imaging," Lab on a Chip 12, 1242 (2012)

    work page 2012

  22. [22]

    Lensfree On-chip Tomographic Microscopy Employing Multi-angle Illumination and Pixel Super-resolution,

    S. O. Isikman, W. Bishara, and A. Ozcan, "Lensfree On-chip Tomographic Microscopy Employing Multi-angle Illumination and Pixel Super-resolution," Journal of Visualized Experiments (2012)

    work page 2012

  23. [23]

    Field -Portable Pixel Super - Resolution Colour Microscope,

    A. Greenbaum, N. Akbari, A. Feizi, W. Luo, and A. Ozcan, "Field -Portable Pixel Super - Resolution Colour Microscope," PLoS ONE 8, e76475 (2013)

    work page 2013

  24. [24]

    Wide -field computational color imaging using pixel super-resolved on-chip microscopy,

    A. Greenbaum, A. Feizi, N. Akbari, and A. Ozcan, "Wide -field computational color imaging using pixel super-resolved on-chip microscopy," Optics Express 21, 12469 (2013)

    work page 2013

  25. [25]

    Increased space-bandwidth product in pixel super -resolved lensfree on -chip microscopy ,

    A. Greenbaum, W. Luo, B. Khademhosseinieh, T. -W. Su, A. F. Coskun, and A. Ozcan, "Increased space-bandwidth product in pixel super -resolved lensfree on -chip microscopy ," Sci. Rep. 3, (2013)

    work page 2013

  26. [26]

    Demosaiced pixel super -resolution for multiplexed holographic color imaging,

    Y. Wu, Y. Zhang, W. Luo, and A. Ozcan, "Demosaiced pixel super -resolution for multiplexed holographic color imaging," Sci Rep 6, (2016)

    work page 2016

  27. [27]

    Image quality assessment: from error visibility to structural similarity,

    Z. Wang, A. C. Bovik, H. R. Sheikh, and E. P. Simoncelli, "Image quality assessment: from error visibility to structural similarity," IEEE transactions on image processing 13, 600–612 (2004)

    work page 2004

  28. [28]

    Comparative analysis of the quantization of color spaces on the basis of the CIELAB color -difference formula,

    B. Hill, T. Roger, and F. W. Vorhagen, "Comparative analysis of the quantization of color spaces on the basis of the CIELAB color -difference formula," ACM Transactions on Graphics (TOG) 16, 109–154 (1997)

    work page 1997

  29. [29]

    J. W. Goodman, Introduction to Fourier Optics (Roberts and Company Publishers, 2005)

    work page 2005

  30. [30]

    Edge sparsity criterion for robust ho l- ographic autofocusing,

    Y. Zhang, H. Wang, Y. Wu, M. Tamamitsu, and A. Ozcan, "Edge sparsity criterion for robust ho l- ographic autofocusing," Optics Letters 42, 3824 (2017)

    work page 2017

  31. [31]

    Comparison of Gini index and Tamura coefficient for holographic autofocusing based on the edge sparsity of the complex optical wavefront

    M. Tamamitsu, Y. Zhang, H. Wang, Y. Wu, and A. Ozcan, "Comparison of Gini index and Tam u- ra coefficient for holographic autofocusing based on the edge sparsity of the complex optical wavefront," arXiv:1708.08055 [physics.optics] (2017)

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  32. [32]

    Generative adversarial nets,

    I. Goodfellow, J. Pouget -Abadie, M. Mirza, B. Xu, D. Warde -Farley, S. Ozair, A. Courville, and Y. Bengio, "Generative adversarial nets," in Advances in Neural Information Processing Systems (2014), pp. 2672–2680

    work page 2014

  33. [33]

    Deep learning enables cross -modality super-resolution in fluorescence microscopy,

    H. Wang, Y. Rivenson, Y. Jin, Z. Wei, R. Gao, H. Günaydın, L. A. Bentolila, C. Kural, and A. O z- can, "Deep learning enables cross -modality super-resolution in fluorescence microscopy," Nature Methods 16, 103–110 (2019)

    work page 2019

  34. [34]

    Deep learning -based super-resolution in coherent imaging systems,

    T. Liu, K. de Haan, Y. Rivenson, Z. Wei, X. Zeng, Y. Zhang, and A. Ozcan, "Deep learning -based super-resolution in coherent imaging systems," Scientific reports 9, 3926 (2019)

    work page 2019

  35. [35]

    Resolution enhancement in sca n- ning electron microscopy using deep learning,

    K. de Haan, Z. S. Ballard, Y. Rivenson, Y. Wu, and A. Ozcan, "Resolution enhancement in sca n- ning electron microscopy using deep learning," arXiv:1901.11094 [physics] (2019)

    work page arXiv 1901

  36. [36]

    Bright -field ho- lography: cross-modality deep learning enables snapshot 3D imaging with bright -field contrast us- ing a single hologram,

    Y. Wu, Y. Luo, G. Chaudhari, Y. Rivenson, A. Calis, K. de Haan, and A. Ozcan, "Bright -field ho- lography: cross-modality deep learning enables snapshot 3D imaging with bright -field contrast us- ing a single hologram," Light: Science & Applications 8, 25 (2019)

    work page 2019

  37. [37]

    Virtual histological staining of unlabelled tissue -autofluorescence images via deep learning,

    Y. Rivenson, H. Wang, Z. Wei, K. de Haan, Y. Zhang, Y. Wu, H. Günaydın, J. E. Zuckerman, T. Chong, and A. E. Sisk, "Virtual histological staining of unlabelled tissue -autofluorescence images via deep learning," Nature Biomedical Engineering 1 (2019)

    work page 2019

  38. [38]

    U-Net: Convolutional Networks for Biomedical Image Segmentation

    O. Ronneberger, P. Fischer, and T. Brox, "U -Net: Convolutional Networks for Biomedical Image Segmentation," arXiv:1505.04597 [cs] (2015)

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  39. [39]

    Phase recovery and holographic image reconstruction using deep learning in neural networks,

    Y. Rivenson, Y. Zhang, H. Güna ydın, D. Teng, and A. Ozcan, "Phase recovery and holographic image reconstruction using deep learning in neural networks," Light: Science & Applications 7, 17141 (2018)

    work page 2018

  40. [40]

    Understanding the difficulty of training deep feedforwa rd neural ne t- works. 2010,

    X. Glorot and Y. Bengio, "Understanding the difficulty of training deep feedforwa rd neural ne t- works. 2010," in Internaional Conference on Artificial Intelligence and Statistics (n.d.)

    work page 2010

  41. [41]

    Adam: A Method for Stochastic Optimization

    D. P. Kingma and J. Ba, "Adam: A method for stochastic optimization," arXiv preprint arXiv:1412.6980 (2014)

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  42. [42]

    Globally optimal stitching of tiled 3D microscopic image acquisitions,

    S. Preibisch, S. Saalfeld, and P. Tomancak, "Globally optimal stitching of tiled 3D microscopic image acquisitions," Bioinformatics 25, 1463–1465 (2009). Figure 1. Comparison between the traditional hyperspectral imaging and the proposed neural network -based approaches for the recon- struction of accurate color images. NH is the number of sample-to-senso...

    work page 2009