H-Net: A Multitask Architecture for Simultaneous 3D Force Estimation and Stereo Semantic Segmentation in Intracardiac Catheters

Javad Dargahi; Mehrdad Zadeh; Pedram Fekri

REVIEW 3 major objections 7 minor 26 references

H-Net: A Multitask Architecture for Simultaneous 3D Force Estimation and Stereo Semantic Segmentation in Intracardiac Catheters

T0 review · 3 major / 7 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read H-Net is claimed to be the first end-to-end architecture that, from two simultaneous X-ray or RGB images of a deflected catheter, outputs both a catheter segmentation for each view and a 3D force vector, matching Y-Net's accuracy without…

desk verdict H-Net is a neat, very small architecture that jointly segments stereo catheter images and estimates 3D tip force, but its clinical claims rest on a synthetic X-ray test that is not yet validated on real fluoroscopy. read the letter →

arxiv 2501.00514 v1 pith:DU36YAZU submitted 2024-12-31 eess.IV cs.AIcs.CVcs.LGcs.RO

Pedram Fekri , Mehrdad Zadeh , Javad Dargahi This is my paper · ORCID

classification eess.IVcs.AIcs.CVcs.LGcs.RO

keywords multitasksegmentationsemanticcatheterforceestimationbiplanefluoroscopyencoder-decoderlightweightarchitecturesyntheticX-raydata

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

H-Net is proposed as the first end-to-end deep learning architecture that performs two tasks at once: segmenting an intracardiac catheter in two simultaneous images (as a biplane fluoroscopy system would produce) and estimating the 3D force at the catheter tip. The authors claim that, unlike prior learning-based force estimators, it needs no separate pre-segmentation step, can resolve force along the z-axis through stereo input, and stays lightweight by sharing parameters between the two sub-networks. Across the paper's RGB and two synthetic X-ray datasets, force estimation matches the established Y-Net baseline ($R^2=0.98$ on RGB and the easier X-ray set, $0.97$ on the harder set), while segmentation achieves mIoU comparable to U-Net with roughly 74 times fewer parameters (0.46M vs 34M). The motivation is practical: real-time force feedback and catheter visualization improve the safety of ablation procedures, where contact forces of 0.1 N to 0.3 N are recommended, and doing both in one cheap model removes the need for sensors in the catheter.

What carries the argument

The load-bearing object is H-Net's twin-sub-network encoder-decoder with shared weights, shaped like the letter H. Two input images pass through separate but weight-shared encoders and decoders; the segmentation heads are $1\times1$ convolutions with sigmoid that classify each pixel as catheter or background. The force head concatenates, across both views, the global-average-pooled bottleneck feature and the global-average-pooled outputs of every decoder block; this vector then goes through dense layers with 64, 32, and 3 units to produce the force estimate. The coupling of decoder embeddings is what lets the regression head see catheter-shape variations from both angles while ignoring background clutter, and weight sharing is what keeps the parameter count at 0.46M. The synthetic X-ray generator, which composites thresholded catheter foregrounds onto random patches cropped from real chest X-rays, provides the training distribution that mimics fluoroscopy appearance.

What would settle it

Run H-Net, trained on the synthetic and RGB data, on real biplane fluoroscopy sequences recorded during an ablation or in a phantom setup with a force-sensing catheter, and compare its force predictions to the sensor readings and its segmentations to manual annotations; if the $R^2$ on real X-rays falls meaningfully below the $0.97$--$0.98$ range reported on synthetic data, or if segmentation mIoU degrades by more than a few points, the synthetic-data assumption is falsified.

Watch

Extended reading notes

Core claim

The central discovery is that a single shared-parameter encoder-decoder can fuse two views of a deflected catheter so that the same feature representations drive both pixel-level segmentation and regression of the applied force vector. H-Net processes two input images through twin sub-networks whose encoder and decoder weights are shared, passes the bottleneck embedding and the decoder embeddings from each side into a central regression head that outputs $x$, $y$, $z$ force components, and uses two $1\times1$-convolution heads to emit segmentation maps. The authors report that this multi-task design reaches $R^2=0.98$ on the RGB and XRay-1 datasets and $R^2=0.97$ on the harder XRay-2 dataset, while the average segmentation mIoU stays at 95.5--95.7, all with only 0.46M trainable parameters. They conclude that this is the first time an end-to-end architecture has been capable of simultaneous stereo catheter segmentation and 3D force estimation.

Load-bearing premise

The load-bearing premise is that synthetic X-ray images, made by pasting a thresholded catheter foreground onto random crops of chest X-rays, look enough like real biplane fluoroscopy that the force and segmentation accuracy measured on them will hold in an operating room.

Editorial extensions

If this is right

A single lightweight model (0.46M parameters) can serve both visualization and contact-force sensing from raw biplane fluoroscopy, so a catheterization setup no longer needs a separate segmentation stage before force estimation.
Force estimation in the z direction, previously missing in monoplane and many learning-based approaches, is achievable from stereo image pairs alone, which is relevant for ablation where the recommended contact-force window is 0.1 N to 0.3 N.
Because the network is trained on synthetic X-ray data and transfers to harder synthetic backgrounds with only a modest error increase (MAE up about 17.9% on XRay-2), the data-generation recipe may reduce the need for large annotated clinical X-ray datasets.
The architecture's low parameter count comes from sharing encoder-decoder weights across views, filter pruning, and omitting normalization layers; this suggests that domain-specific multi-task models can match the segmentation quality of much larger general-purpose networks at a fraction of the cost.
Since the same shared features drive both the segmentation and force heads, the authors' design implies that improvements in one task's feature quality should also tighten the other task's accuracy.

Reading between the lines

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

The synthetic X-ray compositing recipe could be extended to other interventional tools such as guidewires or electrodes; if the compositing assumption holds, the same H-Net structure would likely need little retraining to segment and measure force on those devices.
The error histograms in the paper show a non-zero mean along the z-axis for the harder X-ray set; a natural follow-up is to test whether processing short temporal sequences of fluoroscopy frames reduces this bias, something the paper does not explore.
The paper compares H-Net against Y-Net with pre-segmented input, but not against a strong modern segmentor feeding Y-Net on raw images; a direct comparison would isolate the multitask benefit from the benefit of shared features.
If real biplane fluoroscopy differs from the synthetic distribution in ways such as motion blur, scatter, or overlapping anatomy, the reported performance could degrade; a direct benchmarking study on clinical data would be the decisive next validation.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

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

H-Net is worth a look if you work on catheter tracking or force estimation. The genuinely new thing here is that it is the first end-to-end network that takes two raw images (simulating biplane fluoroscopy) and outputs both segmentation maps and a 3D force vector, with no pre-segmented input required. The architecture is straightforward: shared-weight encoders/decoders with skip connections, plus a regression head fed by global-pooled features from both bottlenecks and decoders. It is also remarkably small (0.46M parameters), far lighter than U-Net (34M) or HR-Net (9.6M), while matching their mIoU on the RGB and the two synthetic X-ray datasets they test on. That is a real engineering achievement, and the ablation of the separate segmentor is a useful contribution.

The methodology is generally careful. The three datasets are clearly described, the training/validation/test split is reported, and the authors show training curves with no obvious overfitting. The synthetic X-ray generator is a reasonable first attempt to create more realistic inputs than plain RGB, and the two difficulty levels (XRay-1 and XRay-2) provide a sensible stress test for background variation.

But there are soft spots, and one is load-bearing. The "state-of-the-art" claim overstates the tables: HR-Net gets slightly higher mIoU (96.1 vs 95.7 on XRay-1), and Y-Net has lower MSE and MAE on RGB (2.8e-05 vs 3.6e-05). Those differences may be within noise, but there are no error bars or multiple-seed runs, so we cannot tell. More importantly, the synthetic X-ray data does not simulate what it claims. The catheter foreground is a clean thresholded silhouette from the RGB setup, composited onto random chest X-ray patches. There is no X-ray projection, attenuation, scatter, motion blur, or detector noise. The segmentation head can therefore succeed by detecting the composited overlay rather than segmenting a catheter under genuine fluoroscopic appearance. The reported R2=0.97-0.98 and mIoU around 95-96 are evidence that the architecture works on this specific synthetic distribution, but they do not establish transfer to real biplane imaging. The conclusion that H-Net improves catheterization safety is unverified until tested on clinical or at least physically realistic X-ray data.

The paper also does not release code or data, which limits reproducibility. For a subfield that already suffers from benchmark fragmentation, that is a minor but real shortcoming.

Who is this for? Researchers in image-guided intervention and robotic catheterization who want a lightweight multitask baseline. It is a legitimate engineering contribution with a clear architecture and honest comparisons, and the shared-parameter design is a sensible idea. If I were editing, I would send it to a serious referee — in fact, it already appeared in RA-L. My own verdict is conditional: the architecture is promising, but the clinical claim needs real-data validation before I would rely on it.

Referee Report

3 major / 7 minor

Summary. The paper proposes H-Net, a multi-input, multi-output encoder-decoder architecture that simultaneously segments a catheter in two views and estimates the 3D contact force from the images. The network uses two weight-shared encoder-decoder branches, two segmentation heads, and one regression head. The authors introduce a synthetic X-ray generator that overlays thresholded catheter foregrounds from a benchtop RGB setup onto chest X-ray patches. They train and evaluate H-Net on one RGB dataset and two synthetic X-ray datasets, and compare force estimation to Y-Net and segmentation to FCN, U-Net, and HR-Net. The reported results suggest that H-Net performs comparably to the baselines while using considerably fewer parameters (0.46M), and it is presented as the first architecture in this line of work to perform both tasks end-to-end.

Significance. If the results hold, H-Net is a meaningful contribution: it demonstrates that simultaneous catheter segmentation and 3D force regression can be learned end-to-end from stereo images, and its weight-sharing design results in a very lightweight model (0.46M parameters) compared to standard segmentation networks. The extension to synthetic X-rays is a useful engineering contribution, and the held-out evaluations on three datasets support the claim that the architecture can perform both tasks on those distributions. The main weaknesses are the unsupported 'state-of-the-art' wording and the lack of validation on real fluoroscopy, which limits the clinical-safety conclusions. The paper does not provide code or trained models, but the architecture and data-generation procedure are described in sufficient detail to be reproducible.

major comments (3)

[Abstract; Section III-B (Tables I and II)] The abstract and conclusion state that H-Net demonstrates 'state-of-the-art performance in both segmentation and force estimation.' This claim is not supported by Tables I and II. In Table II, HR-Net achieves higher mIoU than H-Net on all three datasets (96.1 vs 95.7 on RGB and XRay-1; 95.8 vs 95.5 on XRay-2). In Table I, Y-Net achieves lower MSE (2.8e-05) than H-Net on RGB (3.6e-05) and on XRay-1 (3.3e-05). Thus H-Net is not state-of-the-art in either task on these benchmarks. I recommend rewording the claims to state that H-Net is competitive with the state of the art while offering the unique combination of end-to-end learning and a much smaller parameter count.
[Section II-A; Section IV] The synthetic X-ray generator does not simulate the X-ray imaging process. The foreground is a clean thresholded mask extracted from RGB frames, and the background is random crops of chest X-rays; there is no simulation of projection, attenuation, scatter, motion blur, or detector noise. Consequently, the synthetic images do not reproduce the appearance of a catheter under real fluoroscopy, and the segmentation target is the exact mask used to compose the image. While this is not circular with respect to held-out test sets, it means that the high accuracy on XRay-1 and XRay-2 is evidence only of performance on this synthetic distribution. The abstract and conclusion claim that the work improves catheterization safety; this is a clinical claim that goes beyond the presented evidence. I request (a) a clear statement in the abstract and conclusion that the results are on synthetic X-rays and clinical validation remains open, and (b) either a domain-randomization experiment or a real-fluoroscopy evaluation, or an explicit discussion of the transfer gap and why the current setup is sufficient for the stated claims.
[Section III-B] The comparison claims rest on point estimates with no error bars, no multiple initialization seeds, and no statistical significance tests. The differences between H-Net and HR-Net in Table II are small (e.g., 0.4 percentage points mIoU), and the differences between H-Net and Y-Net in Table I are likewise small (e.g., 0.8e-05 in MSE). With a single run, these differences may be within run-to-run variance. I ask the authors to report the mean and standard deviation over at least three random seeds for the main comparisons, or otherwise justify why a single run is sufficient.

minor comments (7)

[Figure 3 caption] The caption refers to 'X-Net Out' where the network is H-Net; this should be corrected to 'H-Net Out'.
[Section II-A (after Eq. 2)] There is a typo: 'produced by the the global average pooling (1)' should read 'produced by the global average pooling in (1)'.
[Section II-A] 'However, its best to consider' should be 'However, it is best to consider'.
[Table I caption] The abbreviation R/M is used without a prior definition in the text; the caption explains R as RMSE and M as the average maximum force, but this should be stated in the body as well.
[Section III-B, first paragraph] The text says 'FCN and U-Net were randomly initialized and trained on the synthetic XRay-1 and XRay-2 datasets,' but Table II also reports results on RGB. Please clarify that all three datasets were used for training the baselines.
[Section II-B] The terminology for the transposed convolution is inconsistent: the text uses 'transposed convolution' while Figure 2 labels it '2DConv Transpose'. Please unify the terminology.
[General] A dedicated limitations paragraph discussing the single benchtop setup, the synthetic-to-real gap, and the absence of patient variability would improve the manuscript and is strongly recommended.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: force targets are sensor-derived and segmentation is a standard supervised composite task.

full rationale

The derivation chain is not circular. Force supervision comes from a physical force sensor in the benchtop setup described in [13], independent of the images; the network never receives the sensor values as input. Segmentation supervision is the thresholded mask, which for the synthetic X-ray sets is the foreground used to compose the input; this is a standard generative supervision setup, not a hidden-input leak, because the mask is not fed to the network and the model must recover it from the composite image. Held-out test splits are disjoint from training, so the reported R2 and mIoU values reflect generalization within each dataset. The Y-Net comparison cites the authors' prior work, but that prior work provides an external force-sensor ground truth and a published baseline, so the self-citation is not load-bearing in the sense of defining the target. The main validity limitation, namely that synthetic X-rays are composites of thresholded foregrounds over chest X-ray patches and may not transfer to real fluoroscopy, is an external-domain assumption rather than a circularity. No step reduces a prediction to its input by construction.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The central claim rests primarily on the synthetic data pipeline and the prior Y-Net dataset. No new physical entities are postulated; however, several domain assumptions about data realism are load-bearing, and the free parameters are architecture hyperparameters chosen by hand.

free parameters (6)

Number of filters per conv block = 32
Chosen by hand; not tuned on validation.
Number of encoder/decoder blocks = 4
Chosen by hand; matches U-Net depth.
Learning rate = 1e-4
Standard choice; not reported as tuned.
Batch size = 32
Chosen by hand.
Loss weights (beta) = equal
All three loss terms weighted equally; not tuned.
Synthetic background grid size = 16 cells
Chosen by hand for patch composition.

assumptions (4)

domain assumption Synthetic X-Ray images are representative of real fluoroscopy for training and evaluating force and segmentation.
Used to extrapolate results to real operation rooms; never validated against clinical X-Ray data.
domain assumption Thresholded images from the prior Y-Net dataset are accurate segmentation ground truth.
Segmentation targets are derived from thresholding, not clinical annotation, which may introduce label noise.
domain assumption Catheter deflection shape is sufficient to estimate tip forces when the catheter's tendons are not engaged.
Stated in Section II.B; limits the method to non-tendon-actuated conditions.
domain assumption A two-camera geometry approximates biplane fluoroscopy.
Carried from the Y-Net setup [13]; the exact camera calibration and geometry are not detailed.

how reviews work

0 comments

Cite this review

Pith. "Pith review of H-Net: A Multitask Architecture for Simultaneous 3D Force Estimation and Stereo Semantic Segmentation in Intracardiac Catheters." pith.science (2026). https://pith.science/paper/DU36YAZU

@misc{pith2026250100514,
  author       = {Pith},
  title        = {Pith review of: H-Net: A Multitask Architecture for Simultaneous 3D Force Estimation and Stereo Semantic Segmentation in Intracardiac Catheters},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DU36YAZU}},
  note         = {Machine review of arXiv:2501.00514}
}

read the original abstract

The success rate of catheterization procedures is closely linked to the sensory data provided to the surgeon. Vision-based deep learning models can deliver both tactile and visual information in a sensor-free manner, while also being cost-effective to produce. Given the complexity of these models for devices with limited computational resources, research has focused on force estimation and catheter segmentation separately. However, there is a lack of a comprehensive architecture capable of simultaneously segmenting the catheter from two different angles and estimating the applied forces in 3D. To bridge this gap, this work proposes a novel, lightweight, multi-input, multi-output encoder-decoder-based architecture. It is designed to segment the catheter from two points of view and concurrently measure the applied forces in the x, y, and z directions. This network processes two simultaneous X-Ray images, intended to be fed by a biplane fluoroscopy system, showing a catheter's deflection from different angles. It uses two parallel sub-networks with shared parameters to output two segmentation maps corresponding to the inputs. Additionally, it leverages stereo vision to estimate the applied forces at the catheter's tip in 3D. The architecture features two input channels, two classification heads for segmentation, and a regression head for force estimation through a single end-to-end architecture. The output of all heads was assessed and compared with the literature, demonstrating state-of-the-art performance in both segmentation and force estimation. To the best of the authors' knowledge, this is the first time such a model has been proposed

Figures

Figures reproduced from arXiv: 2501.00514 by the authors.

**Figure 2.** The diagram depicts H-Net’s architecture, which includes two sub-networks, each with an encoder and a decoder. All [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** The diagram demonstrates two samples from RGB, [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: The diagram demonstrates the predicted forces in [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

**Figure 5.** Figure 5: The diagram plots the histogram of errors for the output of H-Net’s force estimation head on each dataset in subplot [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 10 canonical work pages

[1]

Y . R. Manda and K. M. Baradhi, Cardiac Catheterization Risks and Complications. StatPearls Publishing, Treasure Island (FL), 2021. [Online]. Available: http://europepmc.org/books/NBK531461

2021
[2]

Complications of pediatric cardiac catheterization: A review in the current era,

R. Mehta, K.-J. Lee, R. Chaturvedi, and L. Benson, “Complications of pediatric cardiac catheterization: A review in the current era,” Catheterization and Cardiovascular Interventions , vol. 72, no. 2, pp. 278–285, 2008. [Online]. Available: https://onlinelibrary.wiley.com/doi/ abs/10.1002/ccd.21580

work page doi:10.1002/ccd.21580 2008
[3]

Safety and effectiveness of a next-generation contact force catheter: Results of the tactisense trial,

M. Y . Lo, P. Sanders, P. Sommer, J. M. Kalman, U. R. Siddiqui, S. Sundaram, C. Piorkowski, N. Olson, S. M. Madej, and D. N. Gibson, “Safety and effectiveness of a next-generation contact force catheter: Results of the tactisense trial,” JACC: Clinical Electrophysiology, 2021

work page 2021
[4]

Safety and efficacy of the thermocool smarttouch surroundflow catheter for atrial fibrillation ablation: A meta-analysis,

C.-F. Chen, X.-F. Gao, M.-J. Liu, C.-L. Jin, and Y .-Z. Xu, “Safety and efficacy of the thermocool smarttouch surroundflow catheter for atrial fibrillation ablation: A meta-analysis,”Clinical cardiology, vol. 43, no. 3, pp. 267–274, 2020

work page 2020
[5]

Deep learning-based haptic guidance for surgical skills transfer,

P. Fekri, J. Dargahi, and M. Zadeh, “Deep learning-based haptic guidance for surgical skills transfer,” Frontiers in Robotics and AI , vol. 7, 2021. [Online]. Available: https://www.frontiersin.org/article/10. 3389/frobt.2020.586707

arXiv 2021
[6]

Towards deep secure tele-surgery,

P. Fekri, P. Setoodeh, F. Khosravian, A. Safavi, and M. H. Zadeh, “Towards deep secure tele-surgery,” in Proceedings of the International Conference on Scientific Computing (CSC) . The Steering Committee of The World Congress in Computer Science, Computer Engineering and Applied Computing (WorldComp), 2018, pp. 81–86

work page 2018
[7]

Real-time contact force measurement: a key parameter for controlling lesion creation with radiofrequency energy,

D. C. Shah and M. Namdar, “Real-time contact force measurement: a key parameter for controlling lesion creation with radiofrequency energy,” Circulation: Arrhythmia and Electrophysiology , vol. 8, no. 3, pp. 713–721, 2015

work page 2015
[8]

Modeling of a steerable catheter based on beam theory,

M. Khoshnam, M. Azizian, and R. V . Patel, “Modeling of a steerable catheter based on beam theory,” in 2012 IEEE International Conference on Robotics and Automation . IEEE, 2012, pp. 4681–4686

work page 2012

Show all 26 references

[9]

A vision-based method for estimating contact forces in intracardiac catheters,

H. Khodashenas, P. Fekri, M. Zadeh, and J. Dargahi, “A vision-based method for estimating contact forces in intracardiac catheters,” in IEEE 1st International Conference on Autonomous Systems, ICAS . IEEE, 2021

2021
[10]

Toward autonomous cardiac catheterization through a parametric finite element simulation with experimental validation,

M. Roshanfar, P. Fekri, and J. Dargahi, “Toward autonomous cardiac catheterization through a parametric finite element simulation with experimental validation,” ICAS 2023, p. 23, 3 2023

2023
[11]

A deep learning force estimator system for intracardiac catheters,

P. Fekri, H. R. Nourani, M. Razban, J. Dargahi, M. Zadeh, and A. Arshi, “A deep learning force estimator system for intracardiac catheters,” in 2021 IEEE International Symposium on Medical Measurements and Applications (MeMeA). IEEE, 2021, pp. 1–6

2021
[12]

A deep learning model for tip force estimation on steerable catheters via learning-from-simulation,

M. Roshanfar, P. Fekri, and J. Dargahi, “A deep learning model for tip force estimation on steerable catheters via learning-from-simulation,” EasyChair, no. 10364, 6 2023

2023
[13]

Y-net: A deep convolutional architecture for 3d estimation of contact forces in intracardiac catheters,

P. Fekri, H. Khodashenas, K. Lachapelle, R. Cecere, M. Zadeh, and J. Dargahi, “Y-net: A deep convolutional architecture for 3d estimation of contact forces in intracardiac catheters,” IEEE Robotics and Automa- tion Letters, vol. 7, no. 2, pp. 3592–3599, 2022

2022
[14]

Catheter segmentation in x-ray fluoroscopy using synthetic data and transfer learning with light u-nets,

M. Gherardini, E. Mazomenos, A. Menciassi, and D. Stoyanov, “Catheter segmentation in x-ray fluoroscopy using synthetic data and transfer learning with light u-nets,” Computer Methods and Programs in Biomedicine , vol. 192, p. 105420, 2020. [Online]. Available: https://www.sci...

2020
[15]

End-to-end real-time catheter segmentation with optical flow-guided warping during endovascular intervention,

A. Nguyen, D. Kundrat, G. Dagnino, W. Chi, M. E. M. K. Abdelaziz, Y . Guo, Y . Ma, T. M. Y . Kwok, C. Riga, and G.-Z. Yang, “End-to-end real-time catheter segmentation with optical flow-guided warping during endovascular intervention,” in 2020 IEEE International Conference on ...

2020
[16]

A hybrid approach to 3d shape estimation of catheters using ultrasound images,

I. Abdulhafiz and F. Janabi-Sharifi, “A hybrid approach to 3d shape estimation of catheters using ultrasound images,” IEEE Robotics and Automation Letters, vol. 8, no. 4, pp. 1912–1919, 2023

1912
[17]

Optimizing ensemble u-net architectures for robust coronary vessel segmentation in angiographic images,

S.-S. Chang, C.-T. Lin, W.-C. Wang, K.-C. Hsu, Y .-L. Wu, C.-H. Liu, and Y . C. Fann, “Optimizing ensemble u-net architectures for robust coronary vessel segmentation in angiographic images,” Scientific Reports, vol. 14, no. 1, p. 6640, Mar 2024. [Online]. Available: https://d...

2024 doi
[18]

High-resolution feature based central venous catheter tip detection network in x-ray images,

Y . Wang, H. K. Lam, Z.-G. Hou, R.-Q. Li, X.-L. Xie, and S.-Q. Liu, “High-resolution feature based central venous catheter tip detection network in x-ray images,” Med Image Anal , vol. 88, p. 102876, Jun. 2023

2023
[19]

Efficient catheter segmentation in 3d cardiac ultrasound using slice-based fcn with deep supervision and f-score loss,

H. Yang, C. Shan, A. F. Kolen, and P. H. N. de With, “Efficient catheter segmentation in 3d cardiac ultrasound using slice-based fcn with deep supervision and f-score loss,” in 2019 IEEE International Conference on Image Processing (ICIP) , 2019, pp. 260–264

2019
[20]

Fully convolutional networks for semantic segmentation,

J. Long, E. Shelhamer, and T. Darrell, “Fully convolutional networks for semantic segmentation,” CoRR, vol. abs/1411.4038, 2014. [Online]. Available: http://arxiv.org/abs/1411.4038

2014 arXiv
[21]

Segnet: A deep con- volutional encoder-decoder architecture for image segmentation,

V . Badrinarayanan, A. Kendall, and R. Cipolla, “Segnet: A deep con- volutional encoder-decoder architecture for image segmentation,” IEEE Transactions on Pattern Analysis and Machine Intelligence , vol. 39, no. 12, pp. 2481–2495, 2017

2017
[22]

Mask R-CNN,

K. He, G. Gkioxari, P. Doll ´ar, and R. B. Girshick, “Mask R-CNN,” CoRR, vol. abs/1703.06870, 2017. [Online]. Available: http://arxiv.org/abs/1703.06870

2017 arXiv
[23]

U-net: Convolutional networks for biomedical image segmentation,

O. Ronneberger, P.Fischer, and T. Brox, “U-net: Convolutional networks for biomedical image segmentation,” in Medical Image Computing and Computer-Assisted Intervention (MICCAI) , ser. LNCS, vol. 9351. Springer, 2015, pp. 234–241, (available on arXiv:1505.04597 [cs.CV]). [Onli...

2015 arXiv
[24]

Segment anything,

A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y . Lo, P. Doll´ar, and R. Girshick, “Segment anything,” arXiv:2304.02643, 2023

2023 arXiv
[25]

Rapid multi- catheter segmentation for magnetic resonance image-guided catheter- based interventions,

A. M. Aleong, A. Berlin, J. Borg, J. Helou, A. Beiki-Ardakani, A. Rink, S. Raman, P. Chung, and R. A. Weersink, “Rapid multi- catheter segmentation for magnetic resonance image-guided catheter- based interventions,” Medical Physics , 2024. [Online]. Available: https://aapm.onl...

2024 doi
[26]

Deep high- resolution representation learning for visual recognition,

J. Wang, K. Sun, T. Cheng, B. Jiang, C. Deng, Y . Zhao, D. Liu, Y . Mu, M. Tan, X. Wang, W. Liu, and B. Xiao, “Deep high- resolution representation learning for visual recognition,” CoRR, vol. abs/1908.07919, 2019. [Online]. Available: http://arxiv.org/abs/1908. 07919 8

1908 arXiv

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.

[1] [1]

Y . R. Manda and K. M. Baradhi, Cardiac Catheterization Risks and Complications. StatPearls Publishing, Treasure Island (FL), 2021. [Online]. Available: http://europepmc.org/books/NBK531461

2021

[2] [2]

Complications of pediatric cardiac catheterization: A review in the current era,

R. Mehta, K.-J. Lee, R. Chaturvedi, and L. Benson, “Complications of pediatric cardiac catheterization: A review in the current era,” Catheterization and Cardiovascular Interventions , vol. 72, no. 2, pp. 278–285, 2008. [Online]. Available: https://onlinelibrary.wiley.com/doi/ abs/10.1002/ccd.21580

work page doi:10.1002/ccd.21580 2008

[3] [3]

Safety and effectiveness of a next-generation contact force catheter: Results of the tactisense trial,

M. Y . Lo, P. Sanders, P. Sommer, J. M. Kalman, U. R. Siddiqui, S. Sundaram, C. Piorkowski, N. Olson, S. M. Madej, and D. N. Gibson, “Safety and effectiveness of a next-generation contact force catheter: Results of the tactisense trial,” JACC: Clinical Electrophysiology, 2021

work page 2021

[4] [4]

Safety and efficacy of the thermocool smarttouch surroundflow catheter for atrial fibrillation ablation: A meta-analysis,

C.-F. Chen, X.-F. Gao, M.-J. Liu, C.-L. Jin, and Y .-Z. Xu, “Safety and efficacy of the thermocool smarttouch surroundflow catheter for atrial fibrillation ablation: A meta-analysis,”Clinical cardiology, vol. 43, no. 3, pp. 267–274, 2020

work page 2020

[5] [5]

Deep learning-based haptic guidance for surgical skills transfer,

P. Fekri, J. Dargahi, and M. Zadeh, “Deep learning-based haptic guidance for surgical skills transfer,” Frontiers in Robotics and AI , vol. 7, 2021. [Online]. Available: https://www.frontiersin.org/article/10. 3389/frobt.2020.586707

arXiv 2021

[6] [6]

Towards deep secure tele-surgery,

P. Fekri, P. Setoodeh, F. Khosravian, A. Safavi, and M. H. Zadeh, “Towards deep secure tele-surgery,” in Proceedings of the International Conference on Scientific Computing (CSC) . The Steering Committee of The World Congress in Computer Science, Computer Engineering and Applied Computing (WorldComp), 2018, pp. 81–86

work page 2018

[7] [7]

Real-time contact force measurement: a key parameter for controlling lesion creation with radiofrequency energy,

D. C. Shah and M. Namdar, “Real-time contact force measurement: a key parameter for controlling lesion creation with radiofrequency energy,” Circulation: Arrhythmia and Electrophysiology , vol. 8, no. 3, pp. 713–721, 2015

work page 2015

[8] [8]

Modeling of a steerable catheter based on beam theory,

M. Khoshnam, M. Azizian, and R. V . Patel, “Modeling of a steerable catheter based on beam theory,” in 2012 IEEE International Conference on Robotics and Automation . IEEE, 2012, pp. 4681–4686

work page 2012