Airway Segmentation Network for Enhanced Tubular Feature Extraction

Qian Zhang; Qibiao Wu; Yagang Wang

REVIEW 4 major objections 4 minor 19 references

Airway Segmentation Network for Enhanced Tubular Feature Extraction

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

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

Pith's one-line read TfeNet claims that rotating convolution kernels, not offset-based deformation, recovers fine airway branches and reports the top ATM22 score of 94.95%.

desk verdict Novel rotation-parameterized kernel with broad benchmark work, but the SOTA headline rests on margins that are within per-case noise. read the letter →

arxiv 2507.06581 v1 pith:P6DKETWF submitted 2025-07-09 eess.IV cs.CV

Qibiao Wu , Yagang Wang , Qian Zhang This is my paper

classification eess.IVcs.CV MSC 68T0792C55

keywords airwaysegmentationdirection-awareconvolutiontubularstructureextractiondeformablefeaturefusioncomputedtomographyclassimbalancetopologypreservation

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

TfeNet is an attempt to fix a specific, practical failure in airway segmentation: standard and deformable convolutions lose thin distal branches, so predicted airway trees come out fragmented. The paper proposes deforming convolution kernels by rotation instead of by learned offsets, splitting each linear kernel at its center and rotating the two arms by learnable angles so the sampling pattern always stays a connected line segment or polyline lying along the airway. Combined with a residual fusion module that mixes x-, y-, and z-direction responses, the network claims the highest overall score (94.95%) on ATM22, the largest airway dataset, and the best topology scores on the BAS benchmark. That matters because distal-airway reachability is the metric that determines whether bronchoscopic navigation and robotic systems can be deployed clinically.

What carries the argument

The load-bearing object is DAConv, a direction-aware convolution: a linear kernel of length $k$ is split at its center and the two arms are each rotated twice (around the y-axis then z-axis for the x-direction kernel) by angles learned per location, with the chain $3\times3\times3$ conv, instance norm, tanh, and scale $q\in(0,\pi]$ turning features into angles bounded by $q=\pi/4$. The rotation matrices $R_x(\theta)$, $R_y(\theta)$, $R_z(\theta)$ define the new sampling positions, so the kernel becomes a bendable line segment or polyline in 3D instead of a fixed straight bar, and trilinear interpolation reads the image at fractional coordinates. Around it sits the TFFM, which fuses DAConv responses from the x, y, and z axes with a standard $3\times3\times3$ branch and preserves the input via residual addition. Training uses the General Union Loss with local-imbalance weights (Tversky loss in ATM22's first stage), patch-based 128$^3$ sampling, and a two-stage Output1 plus Output2 protocol, with nnU-Net serving as Output1 on AIIB23; post-processing keeps the largest connected component and fills holes, which directly supports the reported topology scores.

What would settle it

A controlled replication would settle it: train the top ATM22 entries and TfeNet as single models under an identical protocol, with the same patches, loss, and GPU budget, and compare MeanScore on the same 150-case test split; if the 94.95% figure does not stay on top, or if an offset-based deformable convolution with equal parameter count and a continuity regularizer matches DAConv's TD and BD, then rotation is not the mechanism carrying the result.

Watch

Extended reading notes

Core claim

The paper claims that rotation-based kernel deformation is better than offset-based deformation for tree-like tubular structures, and that this mechanism drives top-ranked airway segmentation. DAConv takes the linear kernels $k\times1\times1$, $1\times k\times1$, and $1\times1\times k$ along the three axes, splits each kernel at its center $(k-1)/2$, and rotates the two halves around the travel axis by angles $\theta_1,\dots,\theta_4$ produced on the fly from the input feature map via a $3\times3\times3$ convolution, instance normalization, tanh, and a scale factor $q=\pi/4$; the resulting fractional sampling positions are read with trilinear interpolation. Because the two arms remain connected through the fixed rotation center, the kernel can bend but never disintegrate into scattered points, which is the failure the paper attributes to deformable convolutions. The Tubular Feature Fusion Module concatenates the three directional DAConv responses with a $3\times3\times3$ context branch and adds the input back through a $1\times1\times1$ residual, preserving faint airway signal. On this machinery the paper reports the best MeanScore (87.37%), tree-length detection (84.27%), and branch detection (77.51%) on BAS; the top DSC (95.49%) and MeanScore (94.95%) on the ATM22 test set; and second place (89.26%) on the fibrosis-focused AIIB23 validation set.

Load-bearing premise

The headline ranking assumes that the published scores of rival methods are fair comparisons, with equal compute and no multi-model ensembling on their side, while TfeNet's numbers come from combining two model outputs and its 0.42-point lead over second place sits well below the 3.90-point case-to-case variation.

Editorial extensions

If this is right

If the mechanism claim holds, any tubular-structure segmentation task, including vessels, coronary arteries, and nerves, can swap offset-based deformable blocks for DAConv blocks and expect more connected predictions, since the paper frames DAConv as generic tubular feature extraction rather than an airway-specific trick.
The reported balance on ATM22, a spread of only 2.2% between the best and worst of the four metrics, means a single TfeNet model needs no accuracy-versus-continuity tuning knob, unlike leaderboard entries that score about 97% on tree length by sacrificing roughly 8 points of precision.
Because TFFM is plug-and-play, existing 3D U-Net-family models can gain branch continuity by replacing convolution blocks rather than redesigning losses or adding a second network.
The ablation's monotone pattern, with $q=\pi/4$ beating $q=\pi/3$ and $q=\pi/2$, implies the benefit comes from restricting the rotation range, so practitioners should keep $q$ small rather than letting angles roam.
On BAS, combining Output1 with Output2 adds 1.99 to 2.66 points of tree-length and branch-detection score over Output1 alone, so the two-stage protocol is itself a large part of the topological gain, a corollary the paper states explicitly.

Reading between the lines

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

My read: DAConv's four learned angles are far fewer free parameters than deformable convolution's per-location offsets, so part of the reported gain may be a low-degree-of-freedom regularization effect; a parameter-matched comparison between rotation and offset designs would separate the geometry from the regularization.
My read: the fixed $q=\pi/4$ was tuned on the BAS test set and carried to ATM22 and AIIB23; fibrosis-distorted airways on AIIB23 may want a different rotation range, so a per-dataset or learned $q$ is a natural extension that could widen or shrink the reported margins.
My read: the ATM22 margin over the runner-up (94.95 vs 94.53) is smaller than the reported per-case standard deviation of 3.90, so the exact leaderboard position is fragile even if the in-house BAS comparison stands; the mechanism claim and the precise ranking should be weighed differently.
My read: the two-stage pipeline, with nnU-Net as Output1 on AIIB23, means the challenge numbers include an ensemble effect; the BAS results, where both TfeNet1 and TfeNet2 come from TfeNet alone, are the cleaner evidence for the DAConv mechanism.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Colleague,

Quick take on arXiv:2507.06581 (TfeNet for airway segmentation). The genuinely new piece is DAConv: a linear convolution kernel whose two halves rotate by learned angles, parameterized as four angles per axis. That is distinct from deformable convolution's free offsets and DSConv's chain offsets, and the rotation algebra in Eq. (5) is internally consistent. The TFFM module is simple—axis-wise concatenation of DAConv features with a 3x3 conv and a residual path—but the ablation (Table 7) shows it helps, and the BAS comparisons against DSConv and deformable baselines are fair and reasonably controlled. Code is public. That's a solid engineering contribution.

The soft spots are about the strength of the claims, not the existence of the method. On the ATM22 test leaderboard, TfeNet's MeanScore is 94.95 vs Timi's 94.53, a 0.42-point margin with per-case SDs around 3.9–4.7 on n=150. By my calculation that's a Welch t ≈ 0.84, p ≈ 0.4. The paper calls this "the highest overall score," which is true as a ranking but not established as a meaningful difference. The BAS margin over nnUNet (1.00 point, SDs 6–8, n=20) and the DAConv-vs-DSConv ablation gains (TD +0.97, BD +1.41, SDs 6–8) sit in the same zone: consistent direction, but within noise. The AIIB23 result is second place, but that pipeline uses nnUNet as Output1 plus TfeNet as Output2—an ensemble compared to single-model leaderboard entries. Also, Eq. (21) as printed defines IOU as 1 - TP/(TP+FP+FN), which is actually 1 - IoU. That's a typo, but it should be fixed.

One thing I'd push back on in the stress-test note: the "peeking" concern. They disclose choosing q=π/4 on the BAS test set and switching to Tversky after seeing ATM22 validation misses; that's overfitting to val/test to some degree, but it's not hidden, and the in-house comparisons are still informative. The bigger issue is statistical: the headline rank is fragile.

Who is this for? People working on tubular structure segmentation or airway annotation pipelines. It's worth a serious referee, because the operator is novel and the evaluation is broad—but the referee should ask for significance tests or paired comparisons, single-stage vs ensemble numbers for the challenge submissions, and a corrected IOU equation. If those come back solid, TfeNet is a useful addition.

Recommendation: send to peer review. Not a desk reject.

Referee Report

4 major / 4 minor

Summary. The manuscript proposes TfeNet, a 3D airway segmentation network combining a novel Direction-Aware Convolution (DAConv) that rotates linear kernel segments via learned angles, and a Tubular Feature Fusion Module (TFFM) that fuses x/y/z DAConv features with a standard convolution via residual connections. The network is evaluated on the BAS, ATM22, and AIIB23 datasets. The paper claims the highest MeanScore (94.95%) on ATM22 test set, best MeanScore on BAS (87.37%), and second overall score on AIIB23 validation (89.26%). The authors also report ablations on BAS comparing DAConv against linear and DSConv constructions, and provide public code.

Significance. If the reported performance differences are real, the paper makes a modest but useful contribution to tubular-structure segmentation: DAConv provides a constrained, interpretable alternative to offset-based deformable convolutions, and the full architecture achieves competitive results on three benchmarks. The strengths are the public code, the disclosure of the loss-function switch after validation inspection, and the controlled in-house BAS comparison with several baselines. However, the headline claims of superiority are not yet supported by the reported statistics: all key comparisons have margins within the per-case standard deviations, and no significance tests are reported. The paper's central claim therefore requires additional evidence or reframing before it can be considered established.

major comments (4)

[§5.2, Table 5] The abstract and Section 5.2 claim that TfeNet 'achieves the highest overall score of 94.95%' on ATM22. The margin over the runner-up (Timi, 94.53) is 0.42 points, while the reported per-case standard deviation is 3.90 for TfeNet and 4.68 for Timi on n=150. This difference is far from significant (approximate Welch t ≈ 0.84, p ≈ 0.40, 95% CI containing 0). Similarly, the BAS results in Table 2 (87.37 vs 86.37, n=20) are within noise. The paper should report appropriate significance tests (paired where applicable, with multiplicity correction) or effect sizes, and the wording should be moderated if the differences are not statistically meaningful.
[§6.1, Table 7] The rotation limit q is selected by comparing DAConv45, DAConv60, and DAConv90 on the BAS test set (Table 7), and this selected value q=π/4 is then used in all experiments, including the BAS results in Table 2. This constitutes test-set tuning and makes the reported BAS superiority non-independent; the same concern applies to the transfer of q to other datasets. The authors should select q on a validation split, or at minimum justify the choice from domain knowledge, and disclose this limitation.
[§6.1, Table 7 and §2.2] The central sub-claim that rotation-based kernels improve over offset-based deformable convolutions is not directly tested. Table 7 compares linear, DSConv, and DAConv within the same TFFM, but DSConv is a continuity-constrained snake convolution, not a standard offset-based deformable convolution. The only offset-based baseline, DCUNet in Table 2, differs in the overall architecture (3D-UNet without TFFM), so the difference cannot be attributed to the kernel parameterization. An ablation arm using classic deformable convolution (with learned offsets) inside the same TFFM is needed to support the stated sub-claim.
[§5.3 and §4.2] The AIIB23 and ATM22 results are obtained with a two-stage pipeline in which nnUNet serves as Output1 and TfeNet as Output2 (Section 4.2; Section 5.3). The leaderboard comparators in Tables 5 and 6 are not necessarily two-stage or ensemble-based, so the comparison may conflate the method's contribution with the benefits of ensembling/refinement. Reporting the Output2-only result (or the single-stage variant of TfeNet) on these datasets would clarify the contribution of DAConv and TFFM.

minor comments (4)

[§4.1, Table 1] The text says the BAS test set contains 'more than 20 scans' while Table 1 lists 20; the numbers should be reconciled.
[§6.1] The sentence 'Compared to Linear, TD and BD increase by 2.% and 2.57%, respectively' contains an incomplete percentage; should read '2.32%' or the correct value.
[§3.1, Eq. (1)] The definition of the sampling locations in Eq. (1) is confusing: the set lists (x−1,1,1), (x−2,1,1), ..., (x+1,1,1), (x+2,1,1), which is not the expected ordered set for a k×1×1 kernel. Please define the coordinates consistently and use proper indexing in Eqs. (4)–(6).
[§2.1] The phrase 'profits from its innovative encoder-decoder structure' should be 'benefits from its innovative encoder-decoder structure'.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: DAConv and TFFM are defined from rotation and fusion equations that are independent of the reported benchmark targets, and the external leaderboard claims are not constructed from the paper's own outputs.

full rationale

The paper's derivation chain is architectural rather than an analytic derivation of the reported scores. DAConv is defined by the rotation matrices in Eqs. (3)-(6) and the angle predictor in Eq. (8); TFFM is defined by concatenation and residual addition in Eqs. (9)-(10); and the losses and metrics (GUL, Tversky, and Eqs. (15)-(22)) are standard external definitions. No equation equates the claimed ATM22, BAS, or AIIB23 results to a fitted parameter of the model, and no 'prediction' is a renamed input. The disclosed tuning choices, namely q=pi/4 selected from the BAS ablation in Table 7 and the switch to Tversky loss after ATM22 validation inspection, are validation/test selection issues rather than circular reductions: q is a fixed hyperparameter, not a fitted value that mathematically determines the reported MeanScore. The external benchmark claims rest on official leaderboard or offline Docker evaluation, and the BAS comparison is against independently reproduced baselines. The statistical weakness of the 0.42-point ATM22 margin relative to the reported per-case standard deviation is a strength-of-evidence concern, not a circularity concern. No load-bearing self-citation chain, imported uniqueness theorem, or ansatz smuggled in via citation is present.

Assumptions & free parameters 6 free parameters · 5 assumptions · 2 invented entities

The contribution is architectural, so the ledger is dominated by design hyperparameters and domain assumptions rather than derived constants. The most consequential hand-fitted choice is q=pi/4, selected on the BAS test set (Table 7) and reused across all three benchmarks. Loss functions (GUL, LIB, Tversky) are imported from the cited literature (Zheng et al. 2020, 2021; Salehi et al. 2017), so the ad hoc burden stays moderate. No new physical entities are postulated; DAConv and TFFM are concrete, implementable operators with public code and ablation evidence.

free parameters (6)

q (DAConv rotation angle limit) = pi/4
Limits the learned rotation angles to [-q, q] (Eq. 8). Chosen after ablation on the BAS test set (Table 7: DAConv45 beats DAConv60 and DAConv90) and reused for all datasets.
Alpha in General Union Loss = 0.05 (Stage 1), 0.1 (Stage 2)
Balance parameter in the adopted GUL (Eq. 11); set separately for the two training stages, with no sensitivity analysis.
Lambda in LIB weight = 0.05
Constant in the local-imbalance weight (Eq. 13); hand-set following Zheng et al. 2021.
Root exponent r in LIB weight = random in [2, 3]
Exponent in Eq. (13); randomly sampled per the text, a stochastic hand-chosen range.
Rotation augmentation threshold = 0.7 (Stage 1), 0.9 (Stage 2)
Hand-chosen augmentation thresholds that differ across stages and datasets (Section 4.2).
DAConv kernel length k = 7
Length of the linear kernels in DAConv and of the comparison deformable kernels; set by hand (Section 4.2, Section 6.1).

assumptions (5)

standard math Rotation matrices (Eq. 3) and trilinear interpolation (Eq. 7) describe DAConv's deformed sampling positions.
Standard 3D rotation and interpolation; Eq. (5) is consistent with the y-then-z rotation order for the x-direction kernel.
domain assumption Airway branching can be represented by rotated linear kernels along the x, y, and z axes, using 'I'-shaped and 'V'-shaped motifs.
Section 3.1 design premise; not independently validated, supported only by the downstream segmentation scores.
ad hoc to paper Four learned angles per kernel segment are sufficient to capture local airway direction.
The angle vector has exactly four entries (Eq. 2); capacity chosen without analysis.
domain assumption GUL with LIB weights (Zheng et al. 2020, 2021) and Tversky loss are appropriate for this task.
Adopted from prior work (Section 3.3) rather than re-derived.
domain assumption Leaderboard baseline numbers (ATM22, AIIB23) are valid comparators produced under comparable protocols.
Tables 4-6 use organizer-reported leaderboard values; the paper does not control the baselines' training data, ensembles, or compute (Sections 5.2, 5.3).

invented entities (2)

Direction-Aware Convolution (DAConv) independent evidence
purpose: Replace standard and deformable convolutions with rotatable linear kernels to focus on tubular airway structure.
A concrete implementable operator; the public code repository and the ablation in Table 7 allow independent verification of its effect.
Tubular Feature Fusion Module (TFFM) independent evidence
purpose: Fuse x/y/z direction-aware features with a 3x3x3 context path and a residual connection to amplify local tubular features.
A plug-and-play module evaluated by ablation (Table 7); independent evidence is the ablation and the public code.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Airway Segmentation Network for Enhanced Tubular Feature Extraction." pith.science (2026). https://pith.science/paper/P6DKETWF

@misc{pith2026250706581,
  author       = {Pith},
  title        = {Pith review of: Airway Segmentation Network for Enhanced Tubular Feature Extraction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/P6DKETWF}},
  note         = {Machine review of arXiv:2507.06581}
}

read the original abstract

Manual annotation of airway regions in computed tomography images is a time-consuming and expertise-dependent task. Automatic airway segmentation is therefore a prerequisite for enabling rapid bronchoscopic navigation and the clinical deployment of bronchoscopic robotic systems. Although convolutional neural network methods have gained considerable attention in airway segmentation, the unique tree-like structure of airways poses challenges for conventional and deformable convolutions, which often fail to focus on fine airway structures, leading to missed segments and discontinuities. To address this issue, this study proposes a novel tubular feature extraction network, named TfeNet. TfeNet introduces a novel direction-aware convolution operation that first applies spatial rotation transformations to adjust the sampling positions of linear convolution kernels. The deformed kernels are then represented as line segments or polylines in 3D space. Furthermore, a tubular feature fusion module (TFFM) is designed based on asymmetric convolution and residual connection strategies, enhancing the network's focus on subtle airway structures. Extensive experiments conducted on one public dataset and two datasets used in airway segmentation challenges demonstrate that the proposed TfeNet achieves more accuracy and continuous airway structure predictions compared with existing methods. In particular, TfeNet achieves the highest overall score of 94.95% on the current largest airway segmentation dataset, Airway Tree Modeling(ATM22), and demonstrates advanced performance on the lung fibrosis dataset(AIIB23). The code is available at https://github.com/QibiaoWu/TfeNet.

Figures

Figures reproduced from arXiv: 2507.06581 by the authors.

**Figure 1.** Application, Motivation and Challenges. (a) Airway segmentation can provide the rendering basis for airway virtual navigation and positioning. (b) Comparison of sampling locations for standard, dilated, and deformable convolutions with direction-aware convolutions. Our method is more capable of focusing on slender structures. (c) Airway segmentation is faced with misdetection and over-segmentation of the airway. bia… view at source ↗

**Figure 2.** This figure illustrates the detailed architecture of the proposed TfeNet. The rotation matrices of a point in space around the x-axis, y-axis, and z-axis are, respectively: ⎧ ⎪ ⎪ ⎪ ⎪ ⎪ ⎨ ⎪ ⎪ ⎪ ⎪ ⎪ ⎩ 𝑅𝑥 (𝜃) = ⎡ ⎢ ⎢ ⎣ 1 0 0 0 cos 𝜃 − sin 𝜃 0 sin 𝜃 cos 𝜃 ⎤ ⎥ ⎥ ⎦ 𝑅𝑦 (𝜃) = ⎡ ⎢ ⎢ ⎣ cos 𝜃 0 sin 𝜃 0 1 0 − sin 𝜃 0 cos 𝜃 ⎤ ⎥ ⎥ ⎦ , 𝑅𝑧 (𝜃) = ⎡ ⎢ ⎢ ⎣ cos 𝜃 − sin 𝜃 0 sin 𝜃 cos 𝜃 0 0 0 1 ⎤ ⎥ ⎥ ⎦ (3) We denote the specific sampling… view at source ↗

**Figure 3.** The detailed flowchart of the DAConv process are as follows: (a) The rotation process of DAConv convolution; (b) Initially, the sampling positions of the convolution kernels are represented by blue, with the rotation center marked by a yellow point. After the first rotation, the positions become orange, and after the second rotation, they turn black; (c) The sampling space of DAConv, where the red, blue, and green l… view at source ↗

Figures from the paper (4 more)

**Figure 4.** Figure 4: illustrates a visual comparison of advanced airway segmentation methods on the BAS test set. The first column displays the ground truth airway labels. The second column shows the segmentation results obtained by our proposed method. The remaining columns present the ai…

**Figure 5.** Figure 5: Analysis of false positive results in airway segmentation on the BAS dataset. Four key regions of false positives were selected. The first row shows the original CT images, and the second row shows the analyzed images. In both the analyzed images and 3D views, false po…

**Figure 6.** Figure 6: Analysis of false negatives results in airway segmentation on the BAS dataset. Four key regions of false negatives were selected. The first row shows the original CT images, and the second row shows the analyzed images. In both the analyzed images and 3D views, false p…

**Figure 7.** Figure 7: DAConv Sampling Locations. The yellow dot represents the center of the selected convolution kernel, while the red dots indicate the sampling positions of the 57 (1+7+49) convolution kernels related to the center kernel. The heatmap shows the overlap degree of these con…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

19 extracted references · 19 canonical work pages

[1]

Thelungimagedatabaseconsortium(lidc) and image database resource initiative (idri): a completed reference database of lung nodules on ct scans

Armato,S.G.,Mclennan,G.,Bidaut,L.,McNitt-Gray,M.F.,Meyer,C.R.,Reeves,A.P.,Zhao,B.,2011. Thelungimagedatabaseconsortium(lidc) and image database resource initiative (idri): a completed reference database of lung nodules on ct scans. Medical physics 38 2, 915–31. URL: https://api.semanticscholar.org/CorpusID:21883326. Azad, R., Niggemeier, L., Huttemann, M....

work page 2011
[4]

Isensee,F.,Jaeger,P.F.,Kohl,S.A.A.,Petersen,J.,Maier-Hein,K.H.,2020

URL:https://api.semanticscholar.org/ CorpusID:221183043. Isensee,F.,Jaeger,P.F.,Kohl,S.A.A.,Petersen,J.,Maier-Hein,K.H.,2020. nnu-net:aself-configuringmethodfordeeplearning-basedbiomedical image segmentation. Nature Methods 18, 203 –

work page 2020
[6]

2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , 6593–6600URL:https: //api.semanticscholar.org/CorpusID:264376617

Landmark based bronchoscope localization for needle insertion under respiratory deformation. 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , 6593–6600URL:https: //api.semanticscholar.org/CorpusID:264376617. He, K., Zhang, X., Ren, S., Sun, J.,

work page 2023
[9]

URL:https://api.semanticscholar.org/CorpusID:41343397

3d convolutional neural networks with graph refinement for airway segmentation using incomplete data labels, in: MLMI@MICCAI. URL:https://api.semanticscholar.org/CorpusID:41343397. Jin,Q.,Meng,Z.P.,Pham,T.D.,Chen,Q.,Wei,L.,Su,R.,2018. Dunet:Adeformablenetworkforretinalvesselsegmentation. Knowl.BasedSyst. 178, 149–162. URL:https://api.semanticscholar.org/C...

work page 2018
[13]

2023 IEEE/CVF International Conference on Computer Vision (ICCV) , 6047–6056URL:https://api.semanticscholar

Dynamic snake convolution based on topological geometric constraints for tubular structure segmentation. 2023 IEEE/CVF International Conference on Computer Vision (ICCV) , 6047–6056URL:https://api.semanticscholar. org/CorpusID:259937654. Qin, Y., Chen, M., Zheng, H., Gu, Y., Shen, M., Yang, J., Huang, X., Zhu, Y.M., Yang, G.Z.,

work page 2023
[14]

URL:https://api.semanticscholar.org/CorpusID:196831594

Airwaynet: A voxel-connectivity aware approachforaccurateairwaysegmentationusingconvolutionalneuralnetworks,in:InternationalConferenceonMedicalImageComputingand Computer-Assisted Intervention. URL:https://api.semanticscholar.org/CorpusID:196831594. Qin, Y., Gu, Y., Zheng, H., Chen, M., Yang, J., Zhu, Y.M., 2020a. Airwaynet-se: A simple-yet-effective appro...

work page 2020
[17]

Artificial intelligence in medicine 143, 102637

Two-stage contextual transformer- based convolutional neural network for airway extraction from ct images. Artificial intelligence in medicine 143, 102637. URL:https: //api.semanticscholar.org/CorpusID:254685958. Yao,T.,Xu,Y.,Huang,H.,Yan,H.,Ma,Y.,2024. Objectre-identificationnetworkbasedondynamicsnakeattentionmodule. 20245thInternational Conference on Co...

work page 2024
[19]

IEEETransactionsonMedicalRoboticsandBionics4,588–598

Robotic-assisted automatic orientation and insertion for bronchoscopy based on imageguidance. IEEETransactionsonMedicalRoboticsandBionics4,588–598. URL: https://api.semanticscholar.org/CorpusID: 251145028. Çiçek,Ö.,Abdulkadir,A.,Lienkamp,S.S.,Brox,T.,Ronneberger,O.,2016. 3du-net:Learningdensevolumetricsegmentationfromsparseannota- tion,in:InternationalCon...

work page 2016

Show all 19 references

[2009]

URL: https://api

Airway segmentation framework for clinical environments. URL: https://api. semanticscholar.org/CorpusID:18882098. Wang,A.,Tam,T.C.C.,Poon,H.,Yu,K.C.,Lee,W.N.,2022. Naviairway:abronchiole-sensitivedeeplearning-basedairwaysegmentationpipeline for planning of navigation bronchosc...

2022 arXiv
[2012]

IEEETransactionsonMedicalImaging31,2093–2107

Extraction of airwaysfromct(exact’09). IEEETransactionsonMedicalImaging31,2093–2107. URL: https://api.semanticscholar.org/CorpusID: 215761541. Milletarì, F., Navab, N., Ahmadi, S.A.,
[2015]

2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , 770–778URL:https://api.semanticscholar.org/CorpusID:206594692

Deep residual learning for image recognition. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , 770–778URL:https://api.semanticscholar.org/CorpusID:206594692. Hofmanninger, J., Prayer, F., Pan, J., Röhrich, S., Prosch, H., Langs, G.,

2016
[2016]

2016 Fourth International Conference on 3D Vision (3DV) , 565–571URL:https://api.semanticscholar.org/CorpusID:206429151

V-net: Fully convolutional neural networks for volumetric medical image segmentation. 2016 Fourth International Conference on 3D Vision (3DV) , 565–571URL:https://api.semanticscholar.org/CorpusID:206429151. : Preprint submitted to Elsevier Page 17 of 19 Nan, Y., Ser, J.D., Tan...

2016
[2017]

2017 IEEE International Conference on Computer Vision (ICCV) , 764–773URL:https://api.semanticscholar.org/CorpusID:4028864

Deformable convolutional networks. 2017 IEEE International Conference on Computer Vision (ICCV) , 764–773URL:https://api.semanticscholar.org/CorpusID:4028864. Ding, X., Guo, Y., Ding, G., Han, J.,

2017
[2018]

2019 International Conference on Robotics and Automation (ICRA) , 5046–5052URL:https://api.semanticscholar.org/CorpusID: 52288689

Offsetnet: Deep learning for localization in the lung using rendered images. 2019 International Conference on Robotics and Automation (ICRA) , 5046–5052URL:https://api.semanticscholar.org/CorpusID: 52288689. Shen, M., Giannarou, S., Yang, G.Z.,

2019
[2019]

2019IEEE/CVFInternationalConferenceonComputerVision(ICCV),1911–1920URL: https://api.semanticscholar.org/CorpusID: 199543841

Acnet: Strengthening the kernel skeletons for powerful cnn via asymmetric convolution blocks. 2019IEEE/CVFInternationalConferenceonComputerVision(ICCV),1911–1920URL: https://api.semanticscholar.org/CorpusID: 199543841. Fried, I., Hoelscher, J., Akulian, J.A., Pizer, S.M., Alte...

1911
[2020]

Neurocomputing 453, 853–864

Adaptive deformable convolutional network. Neurocomputing 453, 853–864. URL: https://api.semanticscholar.org/CorpusID:225273004. Cheng,G.,Wu,X.,Xiang,W.,Guo,C.,Ji,H.,He,L.,2021. Segmentationoftheairwaytreefromchestctusingtinyatrousconvolutionalnetwork. IEEE Access 9, 33583–335...

2021
[2021]

URL:https://api.semanticscholar.org/ CorpusID:237621292

Refined local-imbalance-based weight for airway segmentation in ct, in: International Conference on Medical Image Computing and Computer-Assisted Intervention. URL:https://api.semanticscholar.org/ CorpusID:237621292. : Preprint submitted to Elsevier Page 18 of 19 Zheng,H.,Qin,...

2020
[2022]

IEEE Transactions on Neural Networks and Learning Systems 35, 7391–7404

Fuzzy attention neural network to tackle discontinuity in airway segmentation. IEEE Transactions on Neural Networks and Learning Systems 35, 7391–7404. URL: https://api.semanticscholar.org/CorpusID:252089332. Nan,Y.,Xing,X.,Wang,S.,Tang,Z.,Felder,F.,Zhang,S.,Ledda,R.E.,Ding,X....

2023
[2023]

2024 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) , 1276–1286URL:https://api.semanticscholar.org/CorpusID:261493753

Beyond self-attention: Deformable large kernel attention for medical image segmentation. 2024 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) , 1276–1286URL:https://api.semanticscholar.org/CorpusID:261493753. Chen, F., Wu, F., Xu, J., Gao, G., Ge, Q., Jing, X.,

2024

Pith tools

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

[1] [1]

Thelungimagedatabaseconsortium(lidc) and image database resource initiative (idri): a completed reference database of lung nodules on ct scans

Armato,S.G.,Mclennan,G.,Bidaut,L.,McNitt-Gray,M.F.,Meyer,C.R.,Reeves,A.P.,Zhao,B.,2011. Thelungimagedatabaseconsortium(lidc) and image database resource initiative (idri): a completed reference database of lung nodules on ct scans. Medical physics 38 2, 915–31. URL: https://api.semanticscholar.org/CorpusID:21883326. Azad, R., Niggemeier, L., Huttemann, M....

work page 2011

[2] [4]

Isensee,F.,Jaeger,P.F.,Kohl,S.A.A.,Petersen,J.,Maier-Hein,K.H.,2020

URL:https://api.semanticscholar.org/ CorpusID:221183043. Isensee,F.,Jaeger,P.F.,Kohl,S.A.A.,Petersen,J.,Maier-Hein,K.H.,2020. nnu-net:aself-configuringmethodfordeeplearning-basedbiomedical image segmentation. Nature Methods 18, 203 –

work page 2020

[3] [6]

2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , 6593–6600URL:https: //api.semanticscholar.org/CorpusID:264376617

Landmark based bronchoscope localization for needle insertion under respiratory deformation. 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , 6593–6600URL:https: //api.semanticscholar.org/CorpusID:264376617. He, K., Zhang, X., Ren, S., Sun, J.,

work page 2023

[4] [9]

URL:https://api.semanticscholar.org/CorpusID:41343397

3d convolutional neural networks with graph refinement for airway segmentation using incomplete data labels, in: MLMI@MICCAI. URL:https://api.semanticscholar.org/CorpusID:41343397. Jin,Q.,Meng,Z.P.,Pham,T.D.,Chen,Q.,Wei,L.,Su,R.,2018. Dunet:Adeformablenetworkforretinalvesselsegmentation. Knowl.BasedSyst. 178, 149–162. URL:https://api.semanticscholar.org/C...

work page 2018

[5] [13]

2023 IEEE/CVF International Conference on Computer Vision (ICCV) , 6047–6056URL:https://api.semanticscholar

Dynamic snake convolution based on topological geometric constraints for tubular structure segmentation. 2023 IEEE/CVF International Conference on Computer Vision (ICCV) , 6047–6056URL:https://api.semanticscholar. org/CorpusID:259937654. Qin, Y., Chen, M., Zheng, H., Gu, Y., Shen, M., Yang, J., Huang, X., Zhu, Y.M., Yang, G.Z.,

work page 2023

[6] [14]

URL:https://api.semanticscholar.org/CorpusID:196831594

Airwaynet: A voxel-connectivity aware approachforaccurateairwaysegmentationusingconvolutionalneuralnetworks,in:InternationalConferenceonMedicalImageComputingand Computer-Assisted Intervention. URL:https://api.semanticscholar.org/CorpusID:196831594. Qin, Y., Gu, Y., Zheng, H., Chen, M., Yang, J., Zhu, Y.M., 2020a. Airwaynet-se: A simple-yet-effective appro...

work page 2020

[7] [17]

Artificial intelligence in medicine 143, 102637

Two-stage contextual transformer- based convolutional neural network for airway extraction from ct images. Artificial intelligence in medicine 143, 102637. URL:https: //api.semanticscholar.org/CorpusID:254685958. Yao,T.,Xu,Y.,Huang,H.,Yan,H.,Ma,Y.,2024. Objectre-identificationnetworkbasedondynamicsnakeattentionmodule. 20245thInternational Conference on Co...

work page 2024

[8] [19]

IEEETransactionsonMedicalRoboticsandBionics4,588–598

Robotic-assisted automatic orientation and insertion for bronchoscopy based on imageguidance. IEEETransactionsonMedicalRoboticsandBionics4,588–598. URL: https://api.semanticscholar.org/CorpusID: 251145028. Çiçek,Ö.,Abdulkadir,A.,Lienkamp,S.S.,Brox,T.,Ronneberger,O.,2016. 3du-net:Learningdensevolumetricsegmentationfromsparseannota- tion,in:InternationalCon...

work page 2016