Pith. sign in

REVIEW 3 major objections 5 minor 35 references

A Temporal Convolutional Network-Based Approach and a Benchmark Dataset for Colonoscopy Video Temporal Segmentation

T0 review · 3 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read ColonTCN, a compact temporal convolutional network, outperforms established models on the first open benchmark for segmenting full colonoscopy videos into nine anatomical and procedural-phase classes, while using fewer parameters.

desk verdict Useful first benchmark for colonoscopy video temporal segmentation, but the ColonTCN SOTA claim rests on single-run margins and should be framed as plausible rather than established. read the letter →

arxiv 2502.03430 v2 pith:SECJ2GZ2 submitted 2025-02-05 cs.CV eess.IV

Carlo Biffi , Giorgio Roffo , Pietro Salvagnini , Andrea Cherubini This is my paper · ORCID
classification cs.CVeess.IV
keywords colonoscopyvideotemporalsegmentationconvolutionalnetworkbenchmarkdatasetframe-levelannotationsurgicalworkflowanalysisclassimbalance
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 aims to open up a new task — automatically dividing a full colonoscopy video into anatomical regions and procedural phases — by releasing frame-level annotations for the 60 videos of the open-access REAL-Colon dataset and by proposing ColonTCN, a compact temporal convolutional network for the job. If the approach and benchmark are adopted, automated colonoscopy reporting could compute quality metrics such as withdrawal time, cecal intubation, and bowel-preparation scores directly from video, and researchers could compare models on a common public dataset. The paper's central result is that ColonTCN reaches the highest weighted F1 on both proposed cross-validation splits (76.4 in the 5-fold setting, 62.9 in the 4-fold setting versus 75.2 and 61.0 for ASFormer) with roughly 0.9 million parameters, less than the transformer-based competitor.

What carries the argument

The central object is the ColonTCN temporal block (TB): two weight-normalized, dilated, acausal 1D convolutional layers with kernel size 7 and 64 channels, each followed by ReLU and dropout, wrapped by a residual connection $H_l = \mathrm{ReLU}(H_{l-1} + C_{2,l})$, with dilation factor $2^l$ at block $l$. A $1\times1$ feature-reduction layer first compresses ResNet-50 frame features from 2048 to 64 dimensions, and stacking 12–13 such blocks gives an exponentially growing temporal receptive field spanning the whole procedure. What the block does is let the network see both past and future context across an hour-long video at low parameter cost, which is what the ablations identify as the source of the performance gain.

What would settle it

Re-annotate a random subset of, say, ten REAL-Colon videos with independent gastroenterologists blinded to the released labels and measure frame-level agreement; if Cohen's kappa falls below roughly 0.8, the benchmark's ground truth is too noisy to support the reported model rankings. A weaker check would be to rerun the released code and splits and verify that ColonTCN reproduces wF1=76.4 on the 5-fold protocol, since only the best of four ASFormer sessions was reported.

Watch

Extended reading notes

Core claim

On the paper's terms, the discovery is that a deliberately simple stack of custom temporal convolutional blocks — double dilated, acausal, weight-normalized 1D convolutions with dropout and residual connections, preceded by a feature-reduction layer — is enough to outperform both multi-stage TCNs and a windowed-attention transformer on full-length colonoscopy videos, while using fewer parameters and fewer GFLOPs. The authors further claim that their released annotations for 2.7 million frames constitute the first open benchmark for colonoscopy video temporal segmentation, and that their dual k-fold protocol, particularly the 4-fold leave-one-cohort-out setting, measures how well models generalize to unseen acquisition centers. The ablation studies attribute the gain to the feature-reduction layer, double dilated convolutions, dropout, and residual connections, and show that adding refinement stages hurts rather than helps this task.

Load-bearing premise

The whole benchmark rests on the assumption that the frame-level labels produced by the eight annotators and their gastroenterologist supervisors are accurate and consistent enough to serve as ground truth; the paper reports the annotation protocol and a 0.2% uncertain-frame rate, but gives no inter-annotator agreement numbers.

Editorial extensions

If this is right

  • Automated colonoscopy reports could derive withdrawal time, cecal and ileal intubation rates, and BBPS-related colon segments directly from a single forward pass over the procedure video.
  • The released annotations and k-fold splits give other groups a standard benchmark, so future colonoscopy segmentation methods can be compared on the same 60 videos.
  • The 4-fold leave-one-cohort-out results quantify how much performance drops when models meet unseen centers, setting a realistic expectation for clinical deployment.
  • Because ColonTCN runs acausally offline and the frame encoder can run in real time, the model can be added to existing clinical workflows without modifying the procedure.
  • The finding that multi-stage refinement does not help this task suggests that for long, irregular colonoscopy videos, investing in more temporal blocks is more effective than adding refinement stages.

Reading between the lines

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

  • Editorial inference: the same annotated segments could be used for polyp location reporting, pairing each detected polyp with an anatomical label, which the paper lists as a downstream goal but does not evaluate.
  • Editorial inference: the 4-fold protocol's structure — training on two cohorts and testing on a third — makes it a natural testbed for domain-adaptation methods, since the paper shows all models lose substantial performance in that setting.
  • Editorial inference: because the ileum is absent in 34 of 60 videos, a model's ileum F1 is highly sensitive to the chosen split; future benchmarks may want to report split-level confidence intervals rather than pooled averages.
  • Editorial inference: if the annotations are confirmed by independent re-labeling, the benchmark could be extended to other endoscopic procedures by reusing the same annotation protocol.
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.

Referee Report

3 major / 5 minor

Summary. The paper introduces the first publicly released frame-level annotations for the REAL-Colon dataset (60 full-procedure colonoscopy videos, 2.7M frames), defining a nine-class temporal segmentation task that combines procedural phases (outside, insertion, withdrawal) with seven anatomical colon segments. It proposes ColonTCN, a temporal convolutional network built from custom residual, double-dilated, dropout-regularized temporal blocks, together with an optional multi-stage extension. A dual evaluation protocol is introduced: a 5-fold setting with train/validation/test splits across all cohorts, and a harder 4-fold setting that evaluates on a held-out center. Experiments compare ColonTCN against TeCNO (TCN and MS-TCN) and ASFormer on wF1, wJacc, WMAPE, parameters, and GFLOPs, and include ablations on the proposed block components, number of blocks, multi-stage refinement, augmentation, and loss functions. The paper claims state-of-the-art classification accuracy with low parameter count in both cross-validation settings.

Significance. If validated, the released annotations and benchmark protocol would fill a real gap: there is currently no public frame-level benchmark for colonoscopy temporal segmentation, and this dataset is large (60 full procedures, six centers) and directly downstream of clinically relevant quality metrics (withdrawal time, cecal/ileum intubation, BBPS scoring). The ColonTCN architecture is simple, parameter-efficient, and the ablations show that each proposed component (feature reduction, double dilated residual convolutions, dropout) contributes positively. The paper ships code and data links, and the proposed 4-fold cross-center evaluation is a useful, more challenging protocol. The central relative-performance claim, however, currently rests on single-run comparisons with known training instability in the strongest competitor, and one stated claim (lowest WMAPE) is contradicted by the paper's own table.

major comments (3)
  1. [Section 3.2, Table 2, Discussion] The claim that ColonTCN achieved the lowest WMAPE is not supported by Table 2. In the 5-fold setting, ASFormer has WMAPE=3.3 while ColonTCN has WMAPE=4.3; in the 4-fold setting the values are 16.8 and 16.2 respectively. The Discussion additionally states 'We report a Mean Absolute Percentage Error (MAPE) of 3.1% on withdrawal time for the best model in the 5-fold CV setting', a number that appears in no table. Please correct the claim, identify precisely which model and setting the 'lowest WMAPE' statement refers to, and reconcile the 3.1% figure with Table 2.
  2. [Section 3.2, Table 2] The state-of-the-art claim is not statistically grounded. Table 2 reports a single wF1/wJacc/WMAPE value per model per setting, with no standard deviations, confidence intervals, or per-fold breakdowns. ASFormer is reported as the best of four training sessions 'due to significant training instability', whereas ColonTCN appears to be a single run; comparing a best-of-four result against a single run is not an equal comparison. With test sets of only 12 videos per fold and rare classes (e.g., ileum appears in 26 of 60 videos), the observed wF1 advantage of 1.2–1.9 points can plausibly be within seed or selection noise. Please report per-fold results and variance across at least a few training seeds, and/or paired statistical tests, for all models.
  3. [Section 2.1] For a contribution that is primarily a new benchmark dataset, the paper provides no annotation reliability evidence. Section 2.1 describes eight annotation specialists supervised by gastroenterologists and mentions 'multiple feedback rounds', but reports no inter-annotator agreement (e.g., kappa or per-boundary agreement), no adjudication statistics, and no independent validation of the released labels. All annotators and supervisors are affiliated with Cosmo Intelligent Medical Devices, which also developed ColonTCN and the REAL-Colon dataset. Please provide quantitative annotation reliability metrics and describe the adjudication process; this is needed to establish the benchmark's validity for the community, even though it does not by itself invalidate the relative model comparison.
minor comments (5)
  1. [Section 3.2] The comparison table is referred to as 'Table 8' throughout Sections 3.2 and 4, but the table is labeled 'Table 2' in the manuscript; please unify the numbering.
  2. [Section 3.2] Two consecutive paragraphs in Section 3.2 are verbatim duplicates ('ColonTCN achieved the highest wF1 and wJacc scores and the lowest WMAPE...' appears twice); please remove the duplicate.
  3. [Table 5] In the 5-fold section of Table 5, the first row (Levels 13, Stages 0) lists wF1 as '6.4 65.0 4.3'; this appears to be a typo for '76.4' given the other numbers in the same row and the corresponding Table 2 entry, but as printed it is illegible.
  4. [Table 1] Table 1 column headings list class names without a unit; the body is frame counts, so please state 'frames' in the header and explicitly note that counts are at the 5 fps standardization.
  5. [Section 2.4] WMAPE is defined as the average absolute percentage error of withdrawal-time estimates, but the text in Section 2.4 says 'employing the absolute mean percentage error' and the equation uses per-video absolute percentage errors; please align the wording with the formula.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: ColonTCN is benchmarked against external baselines on newly released annotations, and the self-cited REAL-Colon dataset is open access and not load-bearing for the model derivation.

full rationale

The paper's claimed derivation chain is self-contained: ColonTCN is defined by Eqs. (1)-(3) from ResNet-50 features and standard temporal-convolution components, and its central claim is an empirical comparison with TeCNO, MS-TCN, and ASFormer on test folds constructed from the annotated REAL-Colon videos. The REAL-Colon dataset is cited from the authors' prior work [2], but the dataset itself is open access and the frame-level annotations used here are newly released; no load-bearing mathematical, architectural, or uniqueness claim is justified solely by that citation. The 5-fold and 4-fold splits are described explicitly, model selection is performed on validation data, and test-set numbers are not used as fitted inputs. Concerns about label reliability, single-run comparisons, and the Section 3.2 WMAPE inconsistency are correctness or statistical-rigor issues rather than circularity, because they do not make a reported prediction equivalent to its inputs by construction. No circular step is therefore identified.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim is empirical and rests on unproven premises: the correctness of the new annotations (no inter-annotator agreement reported), the sufficiency of fixed ImageNet features, and the representativeness of 60 multicentric videos. Hyperparameters such as the number of temporal blocks, dropout, and loss weights were selected on validation data, which is standard but means the reported performance is optimistic relative to a fully independent test. No invented entities are introduced.

free parameters (5)
  • Number of temporal blocks (TBs) = 13 (5-fold), 12 (4-fold)
    Chosen by ablation over levels 10-14 (Table 4); affects receptive field and performance.
  • Truncated MSE threshold tau = 4
    Hyperparameter in the loss function (Eq. 6), set after preliminary grid search.
  • Balancing weight lambda = 0.15
    Hyperparameter in the total loss (Eq. 7), set after preliminary grid search.
  • Dropout rate = 0.5
    Dropout after each dilated convolution in the temporal blocks, chosen by the authors; ablation shows it improves wF1.
  • Feature dimension F = 64
    Number of channels in temporal blocks after feature reduction, chosen by the authors.
assumptions (4)
  • standard math The receptive field of the dilated TCN grows exponentially with the number of blocks, enabling full-video context.
    Relies on standard dilated convolution properties; not formally derived in the paper but consistent with TCN literature.
  • domain assumption The nine class labels and the annotation protocol correctly represent the clinical phases and anatomical segments of colonoscopy.
    The benchmark's ground truth is defined by annotators supervised by gastroenterologists; no inter-annotator agreement or independent verification is provided (Section 2.1).
  • domain assumption ImageNet-pretrained ResNet-50 features are sufficient to support the temporal segmentation task.
    Used as a fixed feature extractor without fine-tuning; the authors acknowledge this as a limitation (Section 4).
  • domain assumption The 60 videos from four cohorts are representative of the target clinical population.
    The dataset spans six centers and three continents, but is small (60 videos) and the 4-fold CV shows generalization gaps.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Temporal Convolutional Network-Based Approach and a Benchmark Dataset for Colonoscopy Video Temporal Segmentation." pith.science (2026). https://pith.science/paper/SECJ2GZ2

@misc{pith2026250203430,
  author       = {Pith},
  title        = {Pith review of: A Temporal Convolutional Network-Based Approach and a Benchmark Dataset for Colonoscopy Video Temporal Segmentation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SECJ2GZ2}},
  note         = {Machine review of arXiv:2502.03430}
}
read the original abstract

Following recent advancements in computer-aided detection and diagnosis systems for colonoscopy, the automated reporting of colonoscopy procedures is set to further revolutionize clinical practice. A crucial yet underexplored aspect in the development of these systems is the creation of computer vision models capable of autonomously segmenting full-procedure colonoscopy videos into anatomical sections and procedural phases. In this work, we aim to create the first open-access dataset for this task and propose a state-of-the-art approach, benchmarked against competitive models. We annotated the publicly available REAL-Colon dataset, consisting of 2.7 million frames from 60 complete colonoscopy videos, with frame-level labels for anatomical locations and colonoscopy phases across nine categories. We then present ColonTCN, a learning-based architecture that employs custom temporal convolutional blocks designed to efficiently capture long temporal dependencies for the temporal segmentation of colonoscopy videos. We also propose a dual k-fold cross-validation evaluation protocol for this benchmark, which includes model assessment on unseen, multi-center data.ColonTCN achieves state-of-the-art performance in classification accuracy while maintaining a low parameter count when evaluated using the two proposed k-fold cross-validation settings, outperforming competitive models. We report ablation studies to provide insights into the challenges of this task and highlight the benefits of the custom temporal convolutional blocks, which enhance learning and improve model efficiency. We believe that the proposed open-access benchmark and the ColonTCN approach represent a significant advancement in the temporal segmentation of colonoscopy procedures, fostering further open-access research to address this clinical need.

Figures

Figures reproduced from arXiv: 2502.03430 by the authors.

Figure 1
Figure 1. Center: Distribution of frame labels in the REAL-Colon dataset. On [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Box and whisker plot of the percentage of video length occupied by [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 4
Figure 4. Following this, the network includes a series of [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figures from the paper (4 more)
Figure 3
Figure 3. Figure 3: Overview of the proposed ColonTCN approach for colonoscopy video [PITH_FULL_IMAGE:figures/full_fig_p009_3.png]
Figure 4
Figure 4. Figure 4: Overview of the Temporal Block (TB) used in the proposed ColonTCN [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: Visualization of temporal segmentation on 6 random videos of the [PITH_FULL_IMAGE:figures/full_fig_p018_5.png]
Figure 6
Figure 6. Figure 6: REAL-Colon dataset variability. 25 [PITH_FULL_IMAGE:figures/full_fig_p025_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 31 canonical work pages

  1. [1]

    Berzin, Sravanthi Parasa, Michael B

    Tyler M. Berzin, Sravanthi Parasa, Michael B. Wallace, Seth A. Gross, Alessandro Repici, and Prateek Sharma. Position statement on priorities for artificial intelligence in GI endoscopy: a report by the ASGE Task Force. Gastrointestinal Endoscopy, 92(4):951–959, October 2020

    work page 2020

  2. [2]

    Real-colon: A dataset for developing real-world ai applications in colonoscopy

    Carlo Biffi, Giulio Antonelli, Sebastian Bernhofer, Cesare Hassan, Daizen Hirata, Mineo Iwatate, Andreas Maieron, Pietro Salvagnini, and Andrea Cherubini. Real-colon: A dataset for developing real-world ai applications in colonoscopy. Scientific Data , 11(1):539, 2024

    work page 2024

  3. [3]

    A novel AI device for real-time optical characterization of colorectal polyps

    Carlo Biffi, Pietro Salvagnini, Nhan Ngo Dinh, Cesare Hassan, Prateek Sharma, GI Genius CADx Study Group, and Andrea Cherubini. A novel AI device for real-time optical characterization of colorectal polyps. NPJ digital medicine, 5(1):84, June 2022

    work page 2022

  4. [4]

    Develop- ment and validation of a deep learning-based algorithm for colonoscopy quality assessment

    Yuan-Yen Chang, Pai-Chi Li, Ruey-Feng Chang, Yu-Yao Chang, Siou-Ping Huang, Yang-Yuan Chen, Wen-Yen Chang, and Hsu-Heng Yen. Develop- ment and validation of a deep learning-based algorithm for colonoscopy quality assessment. Surgical Endoscopy, 36(9):6446–6455, 2022. 26

    work page 2022

  5. [5]

    Tecno: Surgical phase recognition with multi-stage temporal convolutional networks

    Tobias Czempiel, Magdalini Paschali, Matthias Keicher, Walter Simson, Hubertus Feussner, Seong Tae Kim, and Nassir Navab. Tecno: Surgical phase recognition with multi-stage temporal convolutional networks. In Medical Image Computing and Computer Assisted Intervention–MICCAI 2020: 23rd International Conference, Lima, Peru, October 4–8, 2020, Pro- ceedings,...

    work page 2020

  6. [6]

    Opera: Attention-regularized transform- ers for surgical phase recognition

    Tobias Czempiel, Magdalini Paschali, Daniel Ostler, Seong Tae Kim, Ben- jamin Busam, and Nassir Navab. Opera: Attention-regularized transform- ers for surgical phase recognition. In Medical Image Computing and Com- puter Assisted Intervention–MICCAI 2021: 24th International Conference, Strasbourg, France, September 27–October 1, 2021, Proceedings, Part IV...

    work page 2021

  7. [7]

    Automated colonoscopy withdrawal phase duration estimation using ce- cum detection and surgical tasks classification

    Thomas De Carvalho, Rawen Kader, Patrick Brandao, Juana Gonz´ alez- Bueno Puyal, Laurence B Lovat, Peter Mountney, and Danail Stoyanov. Automated colonoscopy withdrawal phase duration estimation using ce- cum detection and surgical tasks classification. Biomedical Optics Express, 14(6):2629–2644, 2023

    work page 2023

  8. [8]

    Evelien Dekker and Douglas K. Rex. Advances in CRC Prevention: Screen- ing and Surveillance. Gastroenterology, 154(7):1970–1984, May 2018

    work page 1970

Show all 35 references
  1. [9]

    Deep learning in surgical work- flow analysis: A review of phase and step recognition

    Kubilay Can Demir, Hannah Schieber, TobiasDaniel WeiseRoth, Matthias May, Andreas Maier, and Seung Hee Yang. Deep learning in surgical work- flow analysis: A review of phase and step recognition. IEEE Journal of Biomedical and Health Informatics , 2023

    2023

  2. [10]

    Temporal action segmen- tation: An analysis of modern techniques

    Guodong Ding, Fadime Sener, and Angela Yao. Temporal action segmen- tation: An analysis of modern techniques. IEEE Transactions on Pattern Analysis and Machine Intelligence , 2023

    2023

  3. [11]

    Exploring segment-level semantics for on- line phase recognition from surgical videos

    Xinpeng Ding and Xiaomeng Li. Exploring segment-level semantics for on- line phase recognition from surgical videos. IEEE Transactions on Medical Imaging, 41(11):3309–3319, 2022

    2022

  4. [12]

    Ms-tcn: Multi-stage temporal convolu- tional network for action segmentation

    Yazan Abu Farha and Jurgen Gall. Ms-tcn: Multi-stage temporal convolu- tional network for action segmentation. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 3575–3584, 2019

    2019

  5. [13]

    Development and valida- tion of a three-dimensional deep learning-based system for assessing bowel preparation on colonoscopy video

    Lina Feng, Jiaxin Xu, Xuantao Ji, Liping Chen, Shuai Xing, Bo Liu, Jian Han, Kai Zhao, Junqi Li, Suhong Xia, et al. Development and valida- tion of a three-dimensional deep learning-based system for assessing bowel preparation on colonoscopy video. Frontiers in Medicine, 10, 2023

    2023

  6. [14]

    Trans-svnet: Accurate phase recognition from surgical videos via hybrid embedding aggregation transformer

    Xiaojie Gao, Yueming Jin, Yonghao Long, Qi Dou, and Pheng-Ann Heng. Trans-svnet: Accurate phase recognition from surgical videos via hybrid embedding aggregation transformer. In Medical Image Computing and 27 Computer Assisted Intervention–MICCAI 2021: 24th International Con- ...

    2021

  7. [15]

    Artificial intelligence applied to colonoscopy: Is it time to take a step forward? Cancers, 15(8):2193, 2023

    Antonio Z Gimeno-Garc ´ ıa, Anjara Hern´ andez-P´ erez, David Nicol´ as- P´ erez, and Manuel Hern´ andez-Guerra. Artificial intelligence applied to colonoscopy: Is it time to take a step forward? Cancers, 15(8):2193, 2023

    2023

  8. [16]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition , pages 770–778, 2016

    2016

  9. [17]

    Automatic measurement of quality metrics for colonoscopy videos

    Sae Hwang, JungHwan Oh, JeongKyu Lee, Yu Cao, Wallapak Tavanapong, Danyu Liu, Johnny Wong, and Piet C De Groen. Automatic measurement of quality metrics for colonoscopy videos. In Proceedings of the 13th annual ACM international conference on Multimedia , pages 912–921, 2005

    2005

  10. [18]

    Sv-rcnet: workflow recognition from surgical videos using recurrent convolutional network

    Yueming Jin, Qi Dou, Hao Chen, Lequan Yu, Jing Qin, Chi-Wing Fu, and Pheng-Ann Heng. Sv-rcnet: workflow recognition from surgical videos using recurrent convolutional network. IEEE transactions on medical imaging , 37(5):1114–1126, 2017

    2017

  11. [19]

    Multi-task recurrent convolutional network with corre- lation loss for surgical video analysis

    Yueming Jin, Huaxia Li, Qi Dou, Hao Chen, Jing Qin, Chi-Wing Fu, and Pheng-Ann Heng. Multi-task recurrent convolutional network with corre- lation loss for surgical video analysis. Medical image analysis , 59:101572, 2020

    2020

  12. [20]

    Estimating withdrawal time in colonoscopies

    Liran Katzir, Danny Veikherman, Valentin Dashinsky, Roman Goldenberg, Ilan Shimshoni, Nadav Rabani, Regev Cohen, Ori Kelner, Ehud Rivlin, and Daniel Freedman. Estimating withdrawal time in colonoscopies. In European Conference on Computer Vision , pages 495–512. Springer, 2022

    2022

  13. [21]

    Semantic parsing of colonoscopy videos with multi-label temporal networks

    Ori Kelner, Or Weinstein, Ehud Rivlin, and Roman Goldenberg. Semantic parsing of colonoscopy videos with multi-label temporal networks. In Pro- ceedings of the IEEE/CVF International Conference on Computer Vision , pages 2599–2606, 2023

    2023

  14. [22]

    Temporal convolutional networks for action segmentation and de- tection

    Colin Lea, Michael D Flynn, Rene Vidal, Austin Reiter, and Gregory D Hager. Temporal convolutional networks for action segmentation and de- tection. In proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, pages 156–165, 2017

    2017

  15. [23]

    Temporal deformable residual networks for action segmentation in videos

    Peng Lei and Sinisa Todorovic. Temporal deformable residual networks for action segmentation in videos. In Proceedings of the IEEE conference on computer vision and pattern recognition , pages 6742–6751, 2018

    2018

  16. [24]

    Ms-tcn++: Multi-stage temporal convolutional network for action segmentation

    Shi-Jie Li, Yazan AbuFarha, Yun Liu, Ming-Ming Cheng, and Juergen Gall. Ms-tcn++: Multi-stage temporal convolutional network for action segmentation. IEEE transactions on pattern analysis and machine intelli- gence, 2020. 28

    2020

  17. [25]

    Focal loss for dense object detection

    Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, and Piotr Doll´ ar. Focal loss for dense object detection. In Proceedings of the IEEE interna- tional conference on computer vision , pages 2980–2988, 2017

    2017

  18. [26]

    Automated measurement of quality of mucosa inspection for colonoscopy

    Xuemin Liu, Wallapak Tavanapong, Johnny Wong, JungHwan Oh, and Piet C De Groen. Automated measurement of quality of mucosa inspection for colonoscopy. Procedia Computer Science, 1(1):951–960, 2010

    2010

  19. [27]

    Multi-task temporal convolutional networks for joint recognition of surgical phases and steps in gastric bypass procedures

    Sanat Ramesh, Diego Dall’Alba, Cristians Gonzalez, Tong Yu, Pietro Mascagni, Didier Mutter, Jacques Marescaux, Paolo Fiorini, and Nicolas Padoy. Multi-task temporal convolutional networks for joint recognition of surgical phases and steps in gastric bypass procedures. Internat...

    2021

  20. [28]

    Key quality indicators in colonoscopy

    Douglas K Rex. Key quality indicators in colonoscopy. Gastroenterology Report, 11:goad009, 2023

    2023

  21. [29]

    Quality indicators in colonoscopy: an evolving paradigm

    Rebecca Shine, Andrew Bui, and Adele Burgess. Quality indicators in colonoscopy: an evolving paradigm. ANZ journal of surgery , 90(3):215– 221, 2020

    2020

  22. [30]

    Patel, Alessandro Fugazza, Gaia Pellegatta, Piera Alessia Galtieri, Gianluca Lollo, Silvia Carrara, Andrea Anderloni, Douglas K

    Marco Spadaccini, Andrea Iannone, Roberta Maselli, Matteo Badalamenti, Madhav Desai, Viveksandeep Thoguluva Chandrasekar, Harsh K. Patel, Alessandro Fugazza, Gaia Pellegatta, Piera Alessia Galtieri, Gianluca Lollo, Silvia Carrara, Andrea Anderloni, Douglas K. Rex, Victor Savev...

    2021

  23. [31]

    Artificial intelligence for colonoscopy: Past, present, and future

    Wallapak Tavanapong, JungHwan Oh, Michael A Riegler, Mohammed Khaleel, Bhuvan Mittal, and Piet C De Groen. Artificial intelligence for colonoscopy: Past, present, and future. IEEE journal of biomedical and health informatics, 26(8):3950–3965, 2022

    2022

  24. [32]

    Asformer: Transformer for action segmentation

    Fangqiu Yi, Hongyu Wen, and Tingting Jiang. Asformer: Transformer for action segmentation. arXiv preprint arXiv:2110.08568 , 2021

    2021 arXiv

  25. [33]

    Multi-step validation of a deep learning-based system for the quantification of bowel preparation: a prospective, observational study

    Wei Zhou, Liwen Yao, Huiling Wu, Biqing Zheng, Shan Hu, Lihui Zhang, Xun Li, Chunping He, Zhengqiang Wang, Yanxia Li, et al. Multi-step validation of a deep learning-based system for the quantification of bowel preparation: a prospective, observational study. The Lancet Digita...

    2021

  26. [34]

    Deepphase: surgi- cal phase recognition in cataracts videos

    Odysseas Zisimopoulos, Evangello Flouty, Imanol Luengo, Petros Giata- ganas, Jean Nehme, Andre Chow, and Danail Stoyanov. Deepphase: surgi- cal phase recognition in cataracts videos. In Medical Image Computing and 29 Computer Assisted Intervention–MICCAI 2018: 21st Internation...

    2018

  27. [35]

    Adenoma detec- tion rate and colorectal cancer risk in fecal immunochemical test screening programs: An observational cohort study

    Manuel Zorzi, Giulio Antonelli, Claudio Barbiellini Amidei, Jessica Battag- ello, Bastianello German` a, Flavio Valiante, Stefano Benvenuti, Alberto Tringali, Francesco Bortoluzzi, Erica Cervellin, et al. Adenoma detec- tion rate and colorectal cancer risk in fecal immunochemi...

    2023

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.