pith. sign in

arxiv: 2604.24999 · v1 · submitted 2026-04-27 · 💻 cs.CV · cs.AI

BifDet: A 3D Bifurcation Detection Dataset for Airway-Tree Modeling

Ali Keshavarzi , Quentin Bouniot , Benjamin M. Smith , Elsa Angelini This is my paper

Pith reviewed 2026-05-08 04:14 UTC · model grok-4.3

classification 💻 cs.CV cs.AI
keywords airway bifurcation detection3D object detectionCT scan datasetlung imagingbranch point annotationairway tree modeling
0
0 comments X

The pith

BifDet provides the first public set of 3D bounding-box labels for airway bifurcations in CT scans.

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

The paper presents BifDet, a new annotated dataset drawn from the ATM22 CT cohort, to support automated detection of airway branch points. Each annotation uses a 3D bounding box that encloses both the parent airway and the two daughter branches at each split. The authors argue that the absence of such specialized data has slowed progress on tools for mapping lung structure and locating disease. They supply the full set of scans with these boxes and release baseline results from two 3D object detectors, RetinaNet and DETR, evaluated across different box-size ranges. The work therefore supplies both the labeled resource and an initial performance reference for future research.

Core claim

We introduce BifDet, the first publicly-available dataset specialized for 3D airway bifurcation detection, comprising carefully annotated CT scans from the ATM22 open-access cohort with bifurcation bounding boxes covering the parent and daughter branches. As a use-case, we fine-tune and evaluate RetinaNet and DETR for 3D airway bifurcations detection on CT scans, providing detailed pipelines including preprocessing steps and implementation choices, with results reported over various categories of minimal bounding box sizes.

What carries the argument

BifDet dataset of 3D bounding-box annotations for airway bifurcations, each box enclosing the parent branch and its two daughter branches at the split point.

If this is right

  • Detection models can now be trained and compared directly on a shared, publicly released set of 3D airway bifurcation labels.
  • Performance baselines from RetinaNet and DETR across multiple box-size categories establish reference numbers for future method development.
  • Preprocessing pipelines and training choices described in the paper can be reused or modified for other 3D detection tasks on the same scans.
  • Improved bifurcation detection could support downstream tasks such as airway-tree reconstruction and lesion localization in respiratory imaging.

Where Pith is reading between the lines

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

  • If models trained on BifDet transfer to clinical data, the dataset could become a standard benchmark for airway analysis software.
  • The same bounding-box annotation style might be applied to other branching tubular structures such as blood vessels or bronchial trees in different organs.
  • Extending the dataset with additional labels for airway wall thickness or stenosis could turn BifDet into a multi-task resource.

Load-bearing premise

The manual annotations of bifurcation bounding boxes are accurate, consistent, and sufficient to train generalizable 3D detection models without significant label noise or domain shift from the ATM22 source scans.

What would settle it

A detection model trained solely on BifDet shows substantially lower precision and recall when tested on an independent set of CT scans acquired with different scanners or protocols.

Figures

Figures reproduced from arXiv: 2604.24999 by Ali Keshavarzi, Benjamin M. Smith, Elsa Angelini, Quentin Bouniot.

Figure 1
Figure 1. Figure 1: (a) Lung CT scan with highlighted airway region. (b) 3D rendering of the airway tree with the centers of the view at source ↗
Figure 2
Figure 2. Figure 2: Annotation pipeline. The average annotation time per case is approximately 8 hours (+2 hours of final check) view at source ↗
Figure 3
Figure 3. Figure 3: Qualitative results comparing Deformable DETR (top row) and RetinaNet (bottom row) bifurcation detection view at source ↗
Figure 4
Figure 4. Figure 4: Histogram of bifurcation sizes along the X, Y, and Z directions in voxel units. All plots use fixed bin intervals view at source ↗
Figure 5
Figure 5. Figure 5: Grouped bar plots showing the number of bifurcations per size bin along the X, Y, and Z axes. Each bin view at source ↗
Figure 6
Figure 6. Figure 6: A 2D Bifurcation Model depicting the parent view at source ↗
Figure 7
Figure 7. Figure 7: Per-case violin plots of bifurcation bounding box sizes across the X, Y, and Z axes, shown in (a) physical view at source ↗
Figure 8
Figure 8. Figure 8: Per-case distributions of bifurcation sizes along each spatial axis under varying minimum voxel size thresholds view at source ↗
read the original abstract

Thoracic Computed Tomography (CT) scans offer detailed insights into the intricate branching network of the airway tree, which is essential for understanding various respiratory diseases. Airway bifurcations, where airway branches split, are crucial landmarks for understanding lung physiology, disease mechanisms and lesion localization. Despite the significance of bifurcation analysis, a notable lack of datasets annotated for this task hinders the development of advanced automated specialized detection or segmentation tools. In this paper, we introduce BifDet, the first publicly-available dataset specialized for 3D airway bifurcation detection, filling a critical gap in existing resources. Our dataset comprises carefully annotated CT scans from the ATM22 open-access cohort with bifurcation bounding boxes covering the parent and daughter branches. As a use-case for demonstrating the potential of BifDet, we fine-tune and evaluate RetinaNet and DETR for 3D airway bifurcations detection on CT scans. We provide detailed pipelines, including preprocessing steps and specific implementation design choices. Results are detailed over various categories of minimal bounding box sizes to serve as baseline to benchmark future research.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 2 minor

Summary. The paper introduces BifDet, the first publicly available 3D dataset for airway bifurcation detection derived from ATM22 thoracic CT scans. It provides bounding-box annotations that cover parent and daughter branches at each bifurcation, along with baseline experiments fine-tuning RetinaNet and DETR detectors. Preprocessing pipelines and results stratified by minimal bounding-box size categories are included to benchmark future work in airway-tree modeling for respiratory disease analysis.

Significance. If the annotations are shown to be reliable and consistent, the dataset would fill a documented gap in specialized resources for 3D bifurcation detection, enabling progress on automated tools for lung physiology, disease mechanism studies, and lesion localization. The provision of reproducible baselines and size-stratified evaluation strengthens its utility as a community resource.

major comments (3)
  1. [Dataset Construction] The manuscript provides no description of the annotation protocol, including how bounding boxes were placed to cover parent and daughter branches, the number of annotators involved, inter-annotator agreement statistics, or any validation procedure for the bounding boxes. This information is load-bearing for the central claim that BifDet constitutes a high-quality, usable resource (Dataset Construction section).
  2. [Experiments and Results] No error analysis, failure-case discussion, or assessment of label noise and domain shift from the ATM22 source scans is presented. Without these, the baseline results cannot substantiate claims of generalizability across bifurcation sizes (Experiments and Results sections).
  3. [Introduction] The assertion that BifDet is the first publicly available dataset specialized for this task is not supported by a systematic comparison table or literature review demonstrating the absence of comparable resources; a concrete gap analysis is required to ground the novelty claim (Introduction).
minor comments (2)
  1. [Methods] Clarify the exact preprocessing steps and implementation design choices for the 3D detectors (e.g., patch extraction, normalization, or augmentation strategies) so that the pipelines can be reproduced without ambiguity.
  2. [Results] Report additional quantitative details such as the total number of annotated bifurcations, distribution of bounding-box sizes, and per-category performance metrics with standard deviations.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their thorough review and constructive suggestions. We have revised the manuscript accordingly and provide the following point-by-point responses to the major comments.

read point-by-point responses

  1. Referee: [Dataset Construction] The manuscript provides no description of the annotation protocol, including how bounding boxes were placed to cover parent and daughter branches, the number of annotators involved, inter-annotator agreement statistics, or any validation procedure for the bounding boxes. This information is load-bearing for the central claim that BifDet constitutes a high-quality, usable resource (Dataset Construction section).

    Authors: We agree with the referee that a detailed annotation protocol is crucial for validating the dataset's quality. The original manuscript described the annotations as 'carefully annotated' but lacked specifics. In the revised manuscript, we have added a dedicated subsection in the Dataset Construction section that outlines the annotation protocol. This includes the method for placing bounding boxes to cover the parent and daughter branches, the number of annotators, inter-annotator agreement statistics where applicable, and the validation procedures employed. These details will allow readers to assess the reliability of BifDet. revision: yes

  2. Referee: [Experiments and Results] No error analysis, failure-case discussion, or assessment of label noise and domain shift from the ATM22 source scans is presented. Without these, the baseline results cannot substantiate claims of generalizability across bifurcation sizes (Experiments and Results sections).

    Authors: We acknowledge that the initial submission lacked an in-depth error analysis. To address this, the revised manuscript includes an expanded Experiments and Results section with a new error analysis subsection. This discusses observed failure cases, potential sources of label noise inherited from the ATM22 dataset, and considerations regarding domain shift. The analysis is stratified by bounding box size categories to better support claims of generalizability. revision: yes

  3. Referee: [Introduction] The assertion that BifDet is the first publicly available dataset specialized for this task is not supported by a systematic comparison table or literature review demonstrating the absence of comparable resources; a concrete gap analysis is required to ground the novelty claim (Introduction).

    Authors: We appreciate the referee's point on substantiating the novelty claim. In the revised Introduction, we have incorporated a systematic literature review of existing datasets for airway tree analysis and detection tasks. Additionally, we have added a comparison table that highlights the unique aspects of BifDet, such as its focus on 3D bifurcation bounding boxes, in contrast to prior resources. This provides the required gap analysis to support that BifDet is the first public dataset specialized for 3D airway bifurcation detection. revision: yes

Circularity Check

0 steps flagged

No significant circularity; dataset release with standard baselines

full rationale

The paper introduces BifDet as a new annotated dataset derived from the existing ATM22 cohort, with manual bounding-box labels for airway bifurcations, followed by fine-tuning of off-the-shelf 3D detectors (RetinaNet, DETR) to produce baseline numbers. No equations, fitted parameters, predictions, or derivations are present. The claim of being the 'first publicly-available dataset' is a factual statement about data release and annotation effort, not a result obtained by reducing any internal quantity to itself. No self-citations are invoked to justify uniqueness theorems or ansatzes. The work is therefore self-contained as a resource contribution with no load-bearing circular steps.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

This is a dataset paper with no mathematical derivations; it relies on the pre-existing ATM22 cohort and standard 3D object detection architectures without introducing new free parameters, axioms, or postulated entities.

pith-pipeline@v0.9.0 · 5492 in / 1182 out tokens · 44898 ms · 2026-05-08T04:14:19.577687+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

3 extracted references · 3 canonical work pages

  1. [1]

    Joshua Gawlitza et al

    ISSN 15354970. Joshua Gawlitza et al. Finding the right spot: Where to measure airway parameters in patients with copd. European Journal of Radiology, 104:87–93, 7 2018. ISSN 18727727. Ross Girshick. Fast r-cnn. InProceedings of the IEEE international conference on computer vision, pages 1440– 1448, 2015. Michael W. Graham and William E. Higgins. Optimal ...

    work page 2018

  2. [2]

    Wieying Kuo et al

    ISSN 0021-8502. . Wieying Kuo et al. Diagnosis of bronchiectasis and airway wall thickening in children with cystic fibrosis: Objective airway-artery quantification.European Radiology, 27: 4680–4689, 11 2017. ISSN 14321084. 1344 BifDet T.C. Lee, R.L. Kashyap, and C.N. Chu. Building skeleton models via 3-d medial surface axis thinning algorithms. CVGIP: Gr...

    work page 2017

  3. [3]

    ISBN 9781665409506

    Institute of Electrical and Electronics Engineers Inc., 2021. ISBN 9781665409506. Benjamin M. Smith et al. Human airway branch variation and chronic obstructive pulmonary disease.Proceedings of the National Academy of Sciences of the United States of America, 115:E974–E981, 1 2018. ISSN 10916490. 1345 A. Keshavarzi, 2026 Zimeng Tan, Jianjiang Feng, and Ji...

    work page 2021