pith. sign in

arxiv: 1906.11467 · v1 · pith:GSU5AFA3new · submitted 2019-06-27 · 📡 eess.IV · cs.CV

Abnormal Colon Polyp Image Synthesis Using Conditional Adversarial Networks for Improved Detection Performance

Younghak Shin , Hemin Ali Qadir , Ilangko Balasingham This is my paper

Pith reviewed 2026-05-25 14:34 UTC · model grok-4.3

classification 📡 eess.IV cs.CV
keywords polyp image synthesisconditional adversarial networkscolonoscopypolyp detectiongenerative adversarial networksmedical image generationimage-to-image translation
0
0 comments X

The pith

Conditional adversarial networks generate realistic synthetic polyp images from normal colonoscopy frames, improving automatic detection performance.

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

The paper seeks to overcome the shortage of labeled polyp images for training detection algorithms during colonoscopy procedures. It introduces conditional adversarial networks conditioned on an edge-filtered combined input image to synthesize new polyp examples while keeping the underlying frame structures intact. This lets the system create additional training data starting from normal images that are easier to collect than labeled ones. If the approach holds, detection models can achieve higher accuracy without requiring proportionally more real annotated samples. The authors report both visual quality in the outputs and measurable gains when the synthetics are added to training sets.

Core claim

The authors establish that an edge filtering-based combined input conditioned image enables a conditional adversarial network to generate synthetic polyp images that are qualitatively realistic, preserve the original colonoscopy image structures, and quantitatively improve polyp detection performance when incorporated into training data.

What carries the argument

The edge filtering-based combined input conditioned image supplied to a conditional adversarial network whose generator uses multiple dilated convolutions per encoding stage and convolution-based resizing for upsampling.

If this is right

  • Synthetic polyp images can be produced from readily available normal colonoscopy images rather than requiring additional labeled data.
  • The generated images maintain the original structures of the source frames.
  • Adding the synthetic images to training sets raises polyp detection performance.
  • Dilated convolutions in the generator allow consideration of large receptive fields without excessive feature map contraction.
  • Convolution-based upsampling in decoding layers reduces artifacts in the output images.

Where Pith is reading between the lines

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

  • The same conditioning strategy might transfer to other medical imaging domains where labeled anomalies are scarce but normal scans are abundant.
  • Detection models trained this way could require fewer expert annotations overall.
  • Combining the generated images with traditional augmentations such as rotation or contrast adjustment might yield further gains.

Load-bearing premise

An edge filtering-based combined input conditioned image will enable realistic polyp image generations while maintaining the original structures of the colonoscopy image frames.

What would settle it

Train a polyp detector once on real images only and once on real plus the generated synthetic images, then compare precision-recall or F1 scores on a fixed held-out test set of real colonoscopy frames; the claim fails if the augmented set shows no improvement.

read the original abstract

One of the major obstacles in automatic polyp detection during colonoscopy is the lack of labeled polyp training images. In this paper, we propose a framework of conditional adversarial networks to increase the number of training samples by generating synthetic polyp images. Using a normal binary form of polyp mask which represents only the polyp position as an input conditioned image, realistic polyp image generation is a difficult task in a generative adversarial networks approach. We propose an edge filtering-based combined input conditioned image to train our proposed networks. This enables realistic polyp image generations while maintaining the original structures of the colonoscopy image frames. More importantly, our proposed framework generates synthetic polyp images from normal colonoscopy images which have the advantage of being relatively easy to obtain. The network architecture is based on the use of multiple dilated convolutions in each encoding part of our generator network to consider large receptive fields and avoid many contractions of a feature map size. An image resizing with convolution for upsampling in the decoding layers is considered to prevent artifacts on generated images. We show that the generated polyp images are not only qualitatively realistic but also help to improve polyp detection performance.

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

3 major / 1 minor

Summary. The manuscript proposes a conditional adversarial network (cGAN) framework to synthesize abnormal colon polyp images from normal colonoscopy frames in order to augment scarce labeled training data for polyp detection. Instead of a standard binary polyp mask, the authors condition the generator on an edge-filtering-based combined input (polyp mask plus edge map) and employ dilated convolutions in the encoder plus convolution-based resizing for upsampling in the decoder. They assert that this produces qualitatively realistic images that preserve original tissue structures and, crucially, that the synthetics improve downstream polyp detection performance.

Significance. If the detection-improvement claim is substantiated, the work would offer a practical route to data augmentation using easily obtained normal frames, which is a recognized bottleneck in colonoscopy analysis. The architectural emphasis on large receptive fields via dilated convolutions and artifact reduction via convolution upsampling are constructive design choices that could be reusable. However, the absence of any reported quantitative validation for the headline claim substantially reduces the immediate significance of the contribution.

major comments (3)
  1. [Abstract] Abstract: the assertion that the generated images 'help to improve polyp detection performance' is presented without any quantitative metrics (e.g., AP, F1, or ROC), baseline comparisons, dataset sizes, or evaluation protocol, leaving the central claim without visible supporting evidence.
  2. [Abstract] Abstract / Method description: no ablation against plain binary-mask conditioning or against standard photometric augmentations is described, so it is impossible to attribute any measured gain specifically to the proposed edge-filtered conditioning rather than to increased training-set size or detector overfitting to GAN artifacts.
  3. [Abstract] Abstract: the claim that the edge-filtered input 'enables realistic polyp image generations while maintaining the original structures' is stated without any supporting analysis of boundary consistency, texture preservation, or potential artifacts, which is load-bearing for the downstream detection utility.
minor comments (1)
  1. [Abstract] Abstract: the phrase 'normal binary form of polyp mask' is imprecise; a clearer definition of the exact conditioning channels would aid reproducibility.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments. We agree that the abstract requires additional quantitative support and analysis to substantiate the claims. We address each point below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the assertion that the generated images 'help to improve polyp detection performance' is presented without any quantitative metrics (e.g., AP, F1, or ROC), baseline comparisons, dataset sizes, or evaluation protocol, leaving the central claim without visible supporting evidence.

    Authors: We agree that the abstract should include quantitative evidence. The full manuscript reports detection results using standard metrics on augmented vs. baseline datasets; we will revise the abstract to explicitly state key metrics (e.g., detection AP/F1 improvements), dataset sizes, and the evaluation protocol. revision: yes

  2. Referee: [Abstract] Abstract / Method description: no ablation against plain binary-mask conditioning or against standard photometric augmentations is described, so it is impossible to attribute any measured gain specifically to the proposed edge-filtered conditioning rather than to increased training-set size or detector overfitting to GAN artifacts.

    Authors: This is a fair point. We will add an ablation study in the experiments section that directly compares the proposed edge-filtered conditioning against plain binary-mask inputs and against standard photometric augmentations, reporting the resulting detection performance for each variant. revision: yes

  3. Referee: [Abstract] Abstract: the claim that the edge-filtered input 'enables realistic polyp image generations while maintaining the original structures' is stated without any supporting analysis of boundary consistency, texture preservation, or potential artifacts, which is load-bearing for the downstream detection utility.

    Authors: We accept this criticism. The revision will incorporate supporting analysis, including quantitative boundary-consistency metrics (e.g., polyp-region overlap) and visual/texture-preservation comparisons, to substantiate the structural-maintenance claim. revision: yes

Circularity Check

0 steps flagged

No significant circularity; empirical GAN pipeline evaluated on downstream task

full rationale

The paper describes a conditional adversarial network architecture with a proposed edge-filtering combined conditioning input. Claims of realistic generation and improved polyp detection rest on empirical outputs and a downstream detector evaluation rather than any derivation, equation, or self-citation that reduces the result to its own inputs by construction. No self-definitional steps, fitted inputs renamed as predictions, or load-bearing self-citations appear in the abstract or method description. The work is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 2 axioms · 0 invented entities

The approach rests on standard assumptions of conditional GAN convergence and the domain-specific effectiveness of edge filtering for structure preservation; no free parameters or invented entities are explicitly introduced beyond typical neural network training choices.

free parameters (1)
  • GAN training hyperparameters
    Learning rates, batch sizes, and loss weights are chosen during training to achieve realistic generation, as is standard in any neural network method.
axioms (2)
  • standard math Conditional GAN minimax optimization can produce realistic images when given appropriately structured conditioned inputs.
    Core assumption underlying all cGAN image synthesis work.
  • domain assumption Edge filtering on binary masks supplies structural cues that preserve colonoscopy frame realism during generation.
    Specific premise required for the proposed input conditioning to succeed.

pith-pipeline@v0.9.0 · 5728 in / 1336 out tokens · 33491 ms · 2026-05-25T14:34:31.234208+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

38 extracted references · 38 canonical work pages · 4 internal anchors

  1. [1]

    Cancer statistics 2017,

    R. L. Siegel, K. D. Miller and A. Jemal, “Cancer statistics 2017,” CA Cancer J Clin., vol. 67, pp. 7-30, 2017

    work page 2017

  2. [2]

    High -grade dysplasia and invasive carcinoma in colorectal adenomas: a multivariate analysis of the impact of adenoma and patient characteristics,

    M. Gschwantler, S. Kriwanek, E. Lan gner, B. Göritzer, C. Schrutka - Kölbl, E. Brownstone, H. Feichtinger and W. Weiss, “High -grade dysplasia and invasive carcinoma in colorectal adenomas: a multivariate analysis of the impact of adenoma and patient characteristics,” Eur. J. Gastroenterol. Hep atol., vol. 14, no. 2, pp. 183–188, 2002

    work page 2002

  3. [3]

    Polyp detection in colonoscopy videos using deeply-learned hierarchical features,

    S. Park, M. Lee, and N. Kwak, “Polyp detection in colonoscopy videos using deeply-learned hierarchical features,” Seoul Nat. Univ., 2015

    work page 2015

  4. [4]

    Colonoscopic polyp detection using convolutional neural networks,

    S. Park and D. Sargent, “Colonoscopic polyp detection using convolutional neural networks,” SPIE Med. Imag., p. 978528, 2016

    work page 2016

  5. [5]

    Comparative Validation of Polyp Detection Methods in Video Colonoscopy: Results from the MICCAI 2015 Endoscopic Vision Challenge,

    J. Bernal, N. Tajkbaksh,, F. J. Sánchez, J. Matuszewski, H. Chen, L. Yu, Q. Angermann, O. Romain, B. Rustad, I. Balasingham, K. Pogorelov, S. Choi, Q. Debard, L. M. Hen, S. Speidel, D. Stoyanov, P. Brandao, H. Cordova, C. S. Montes, S. R. Gurudu, G. F. Esparrach, X. Dray, J. Liang and A. Histace, "Comparative Validation of Polyp Detection Methods in Video...

    work page 2015

  6. [6]

    Convolutional neural networks for medical image analysis: Full training or fine tuning?

    N. Tajbakhsh, J. Y. Shin, S. R. Gurudu, R. T. Hurst, C. B. Kendall, M. B. Gotway and Jianming Liang, “Convolutional neural networks for medical image analysis: Full training or fine tuning?” IEEE Trans. Med. Imag., vol. 35, no. 5, pp. 1299–1312, May 2016

    work page 2016

  7. [7]

    Integrating online and offline 3D deep learning for automated polyp detection in colonoscopy videos,

    L. Yu, H. Chen, Q. Dou, J. Qin and P. A. Heng, “Integrating online and offline 3D deep learning for automated polyp detection in colonoscopy videos,” IEEE J. Biomed. Health Inform., vol. 21, no.1, pp.65-75, 2017

    work page 2017

  8. [8]

    Goodfellow et al., ‘‘Generative adversarial nets,’’ in Proc

    I. Goodfellow et al., ‘‘Generative adversarial nets,’’ in Proc. Adv. Neural Inf. Process. Syst., pp. 2672–2680, 2014

    work page 2014

  9. [9]

    Gauthier

    J. Gauthier. Conditional generative adversarial networks for convolutional face generation. Technical report, 2015

    work page 2015

  10. [10]

    S. Reed, Z. Akata, X. Yan, L. Logeswaran, B. Schiele, and H. Lee. Generative adversarial text -to-image synthesis. In Proceedings of The 33rd International Conference on Machine Learning, 2016

    work page 2016

  11. [11]

    Li and M

    C. Li and M. Wand. Precomputed real -time texture synthesis with markovian generative adversarial networks. In European Conference on Computer Vision (ECCV), 2016. Author Name: Preparation of Papers for IEEE Access (February 2017) VOLUME XX, 2017

    work page 2016

  12. [12]

    Isola, J

    P. Isola, J. Zhu, T. Zhou, and A. A. Efros. Image -to-image translation with conditional adversarial networks. In Conference on Computer Vision and Pattern Recognition (CVPR), 2017

    work page 2017

  13. [13]

    Towards Adversarial Retinal Image Synthesis

    P. Costa, A. Galdran, M. I. Meyer, M. D. Abràmoff, M. Niemeijer, A. M. Mendonca, and A. Campilho. Towards adversarial retinal image synthesis. arXiv preprint arXiv:1701.08974, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  14. [14]

    Road detection from remote sensing images by generative adversarial networks

    Q Shi, X Liu and X Li, “Road detection from remote sensing images by generative adversarial networks”. IEEE access, vol. 6, pp.25486 - 25494, 2018

    work page 2018

  15. [15]

    Unsupervised representation learning with deep convolutional generative adversarial networks

    A. Radford, L. Metz, S. Chintala, "Unsupervised representation learning with deep convolutional generative adversarial networks", In International Conference on Learning Representat ions (ICLR), 2016

    work page 2016

  16. [16]

    U-net: Convolutional networks for biomedical image segmentation

    O. Ronneberger, P. Fischer, T. Brox, "U-net: Convolutional networks for biomedical image segmentation", Proc. Int. Conf. Medical Image Comput. Comput.-Assisted Intervention, pp. 234-241, 2015

    work page 2015

  17. [17]

    Singing voice separation with deep U-Net convolutional networks,

    Andreas Jansson, Eric J. Humphrey, Nicola Mo ntecchio, Rachel Bittner, Aparna Kumar, and Tillman Weyde, “Singing voice separation with deep U-Net convolutional networks,” in Proceedings of the International Society for Music Information Retrieval Conference (ISMIR), pp.323–332, 2017

    work page 2017

  18. [18]

    Photo-Realistic Single Image Super-Resolution Using a Generative Adversarial Network

    C. Ledig, L. The is, F. Huszár, J. Caballero, A. Aitken, A. Te -jani, J. Totz, Z. Wang, and W. Shi. Photo -realistic single image super - resolution using a generative adversarial network . arXiv preprint arXiv:1609.04802, 2016

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  19. [19]

    Multi -scale context aggrega tion by dilated convolutions

    F. Yu and V. Koltun. “Multi -scale context aggrega tion by dilated convolutions”. In International Conference on Learning Representations (ICLR), 2016

    work page 2016

  20. [20]

    Enhancenet: Single image super -resolution through automated texture synthesis

    M.S.M. Sajjadi, B. Schölkopf and M. Hirsch, “Enhancenet: Single image super -resolution through automated texture synthesis”, In International Conference on Computer Vision (ICCV), 2017

    work page 2017

  21. [21]

    Odena, V

    A. Odena, V. Dumoulin and C. Olah, Deconvolution and checkerboard artifacts, 2016, [online] Available: http://distill.pub/2016/deconvchecke rboard/

    work page 2016

  22. [22]

    R -CNN: Towards real-time object det ection with region proposal networks,

    S. Ren, K. He, R. Girshick, and J. Sun, “R -CNN: Towards real-time object det ection with region proposal networks,” in Advances in Neural Information Processing Systems , Montreal, QC, pp. 91 –99, 2015

    work page 2015

  23. [23]

    Is faster r-cnn doing well for pedestrian detection?,

    L. Zhang, L. Lin, X. Liang, and K. He. “Is faster r-cnn doing well for pedestrian detection?,” In European Conference on Computer Vision (ECCV), pp. 443-457, 2016

    work page 2016

  24. [24]

    Speed/accuracy trade -offs for modern convolutional object detectors

    J. Huang, V. Rathod, C. Sun, M. Zhu, A. Korattikara, A. Fathi, I. Fischer, Z. Wojna, Y. Song, S. Guadarrama and K. Murphy, “Speed/accuracy trade -offs for modern convolutional object detectors”, in Proc. IEEE Conf. on Compute r Vision and Pattern Recognition (CVPR), 2017

    work page 2017

  25. [25]

    Context encoders: Feature learning by inpainting

    D. Pathak, P. Krahenbuhl, J. Donahue, T. Darrell, and A. A. Efros. “Context encoders: Feature learning by inpainting”. In Conference on Computer Vision and Pattern Recognition (CVPR), 2016

    work page 2016

  26. [26]

    DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs

    L.-C. Chen, G. Papa ndreou, I. Kokkinos, K. Murphy, and A. L. Yuille. “DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs”. arXiv preprint arXiv:1606.00915, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  27. [27]

    Visualizing and unders tanding convolutional neural networks

    M. D. Zeiler and R. Fergus. “Visualizing and unders tanding convolutional neural networks”. In European Conference on Computer Vision (ECCV), 2014

    work page 2014

  28. [28]

    Fast and accurate deep network learning by exponential linear units (ELUs)

    D. Clevert, T. Unterthiner and S. Hochreiter, "Fast and accurate deep network learning by exponential linear units (ELUs)", Proc. Int. Conf. Learn. Represent., pp. 1-14, 2016

    work page 2016

  29. [29]

    A Computational Approach to Edge Detection,

    J. Canny, "A Computational Approach to Edge Detection," IEEE Trans. Pattern Anal. Mach. Intell.,vol.8, No. 6, pp. 679-698, 1986

    work page 1986

  30. [30]

    Jung [online] Available: https://github.com/aleju/imgaug

    A. Jung [online] Available: https://github.com/aleju/imgaug

    work page

  31. [31]

    Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning

    C. Szegedy, S. Ioffe, and V. Vanhoucke, “Inception -v4, inception - resnet and the impact of residual connections on learning”, arXiv:1602.07261, 2016

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  32. [32]

    Microsoft COC O: Common objects in context

    T.-Y. Lin, M. Maire, S. Belongie, J. Hays, P. Perona, D. Ramanan, P. Dollar, and C. L. Zitnick. “Microsoft COC O: Common objects in context”, In European Conference on Computer Vision (ECCV) , 2014

    work page 2014

  33. [33]

    Adam: A method for stochastic optimization

    D. Kingma and J. Ba. “Adam: A method for stochastic optimization”. In International Conference on Learning Representations (ICLR), 2015

    work page 2015

  34. [34]

    Imagenet classification with deep convolutional neural networks

    Alex Krizhevsky, Ilya Sutskever, and Geoffrey E Hinton. “Imagenet classification with deep convolutional neural networks”, In Neural Information Processing Systems (NIPS), 2012

    work page 2012

  35. [35]

    Wm -dova maps for accurate polyp h ighlighting in colonoscopy: Validation vs. saliency maps from physicians,

    J. Bernal, J. Snchez, G. F. -Esparrach, D. Gil, C. Rodriguez and F. Vilario, “Wm -dova maps for accurate polyp h ighlighting in colonoscopy: Validation vs. saliency maps from physicians,” Comput. Med. Imag. Graph., vol. 43, pp. 99–111, 2015

    work page 2015

  36. [36]

    Towards Real -Time Polyp Detection in Colonoscopy Videos: Adapting Still Frame -Based Methodologies for Video Sequences Analysis

    Q. Angermann, J. Bernal, C. Sánchez -Montes, M. Hammami, G. Fernández-Esparrach, X. Dray, O. Romain, F. J. Sánchez and A. Histace, “Towards Real -Time Polyp Detection in Colonoscopy Videos: Adapting Still Frame -Based Methodologies for Video Sequences Analysis” In Computer Assisted and Robotic Endoscopy and Clinical Image -Based Procedures, Springer, Cham...

    work page 2017

  37. [37]

    Toward multimodal image -to-image translation,

    J.-Y. Zhu, R. Zhang, D. Pathak, T. Darrell, A. A. Efros, O. Wang, and E. Shechtman. “Toward multimodal image -to-image translation,” In Advances in Neural Information Processing Systems (NIPS), pp. 465–476, 2017

    work page 2017

  38. [38]

    High-resolution image synthesis and semantic manipulation with conditional GANs

    T.-C. Wang, M. -Y. Liu, J. -Y. Zhu, A. Tao, J. Kautz, B. Catanzaro, "High-resolution image synthesis and semantic manipulation with conditional GANs", Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), pp. 1-13, 2018

    work page 2018