List of Papers Browse by Subject Areas Author List
Abstract
Robotic-assisted bronchoscopy enables physicians to access pulmonary lesions with enhanced stability and precision; however, navigation accuracy can be compromised by CT-to-body divergence due to respiratory motion, anatomical deformation during the procedure and tracking system errors. Cone-beam computed tomography (CBCT) has therefore emerged as a key intra-operative imaging modality for biopsy verification and navigation updates. A critical prerequisite for integrating CBCT into the workflow is accurate registration between the navigation system and the CBCT coordinate frame, which relies on precise bronchoscope pose estimation. Existing semi-automated solutions require manual scope tip localization and potential correction of pose parameters, introducing variability and prolonging procedure time. In this work, we propose a fully automated pipeline that estimates the bronchoscope pose directly from CBCT images, which consists of: (1) a two-stage model inferencing strategy for rapid and accurate scope tip segmentation, and (2) a context-aware robust pose estimation algorithm. Evaluated on 25 diverse real-world clinical cases, the method achieved a mean position error of 0.55±0.28mm and orientation error of 2.74±1.47°, with an average processing time of 4.47s, approximately 20x faster than the current approach. This combination of accuracy and efficiency enables safer and more reliable navigation updates while reducing clinician workload and minimizing user-induced errors.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/0704_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{ZhaDin_Automated_MICCAI2026,
author = { Zhang, Dingzhong AND Rafii-Tari, Hedyeh AND Matinfar, Babak},
title = { { Automated Scope Pose Estimation in Real-World Robotic-Assisted Bronchoscopy with Integrated C-Arm Imaging } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16893},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper presents a fully automated pipeline for bronchoscope pose estimation from intra-operative CBCT images in robotic-assisted bronchoscopy. Its main contribution is the integration of a two-stage segmentation strategy with a context-aware pose estimation approach to accurately and efficiently recover both the position and orientation of the bronchoscope without manual intervention. By combining coarse-to-fine U-Net segmentation, a sampling strategy based on intensity differences to determine the heading direction, and CAD model alignment for position refinement, the system achieves very high spatial accuracy while significantly reducing computation time. This work advances the clinical workflow by enabling fast, reliable, and operator-independent registration between CBCT and navigation systems, improving the feasibility of real-time navigation updates in interventional procedures.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
1.Strong clinical relevance and clear problem formulation:The work targets a well-recognized limitation in robotic-assisted bronchoscopy, namely CT-to-body divergence and the lack of reliable intra-operative pose estimation for CBCT-based navigation updates. By directly addressing the need for automated scope localization within the clinical workflow, the paper goes beyond algorithmic novelty and tackles a high-impact translational problem. This focus on workflow integration and reduction of operator dependency is particularly valuable for interventional imaging applications. 2.Well designed and coherent end-to-end pipeline:The method combines a two-stage segmentation strategy with a context-aware pose estimation scheme in a logically consistent manner. The coarse-to-fine segmentation improves robustness and efficiency, while the use of intensity differences to resolve directional ambiguity and CAD model alignment for position refinement shows a thoughtful integration of image cues and prior geometric knowledge. Although individual components are not entirely new, their combination is well-motivated and tailored to the challenges of CBCT data. 3.Demonstration on real-world clinical data with practical performance:The approach is evaluated on real intra-operative CBCT scans collected from clinical procedures, which strengthens its credibility compared to studies limited to simulated or curated datasets. The reported results indicate both high accuracy and substantial speed improvements, enabling near real-time performance. This balance between accuracy and efficiency provides convincing evidence of clinical feasibility and potential for deployment in real navigation systems.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
1.Missing comparison with existing methods:The paper does not include quantitative comparisons with prior pose estimation or registration approaches in bronchoscopy or related interventional imaging settings. Instead, it only evaluates against internal baselines such as single stage inference. This makes it difficult to judge whether the proposed method provides a meaningful improvement over existing clinical or research solutions. 2.Limited methodological novelty: The proposed pipeline mainly integrates well-established components, including two-stage U-Net segmentation, PCA-based axis estimation, heuristic intensity-based direction disambiguation, and rigid alignment with a CAD model. While the combination is practical and well-] engineered, the work does not introduce a fundamentally new formulation or learning strategy, and the contribution is therefore more incremental in nature. 3.Lack of direct validation of clinical impact: Although the method is motivated by improving navigation accuracy and diagnostic yield, the evaluation is limited to segmentation metrics and pose estimation errors. The paper does not demonstrate how these improvements translate into downstream clinical outcomes, such as reduced target registration error or improved tool-in-lesion rate. As a result, the claimed clinical benefit is only indirectly supported.
- Please rate the clarity and organization of this paper
Satisfactory
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not mention open access to source code or data, but provides a clear and detailed description of the algorithm to ensure reproducibility.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
no
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(4) Weak Accept — marginally above the acceptance threshold, but would not mind if rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The paper addresses a clinically important problem in robotic-assisted bronchoscopy and presents a practical, well-integrated solution for fully automated bronchoscope pose estimation from CBCT images, with convincing results on real intra operative data demonstrating strong accuracy and significant runtime improvement, which supports its potential for real world deployment; however, the work lacks comparison with existing methods, relies largely on established components with limited methodological novelty, and does not directly validate improvements in downstream clinical outcomes, so while the system-level contribution and clinical relevance are strong, the experimental completeness and novelty are somewhat limited, leading me to a weak accept recommendation.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Accept
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #2
- Please describe the contribution of the paper
CBCT can be used to help bridge pre-op to intra-op imaging to facilitate navigation for robotic assisted bronchoscopy. This manuscript proposes a method to sequentially segment and determine the orientation of the bronchoscope tip in CBCT images using a UNet followed by analysis of voxel intensity along scope tip. The method was evaluated on imaged recorded during RAB procedures.
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
-The authors employ a simple and likely easily implementable and fast method to identify the tip of a bronchoscope in CBCT. -Bronchoscope/catheter orientation detection is a clinically relevant problem.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
-The novelty of the manuscript is limited. Although bronchoscope tip segmentation is not often reported on in CBCT, there are numerous existing works on cathether/guidewire/endoscope/bronchoscope tip detection/segmentation/orientation-estimation in 3D imaging using UNet-like architectures. I do not know if the pose estimation methodology is novel or not but in general tip pose follows fairly trivially from a well segmented cathether/scope. -The authors focus on bronchoscope navigation in this manuscript. Although scope orientation estimation during navigation is in general a clinically relevant problem, I am not sure that it is most relevant for navigational CBCT. The main challenge of navigational CBCT is the CT-body anatomical divergence which the method proposed here does not address. Device orientation in CBCT generally does not require computational assistance as the devices are designed to be highly radiopaque. This problem is generally approached using fluoroscopy or endoscopic video as an input, with CBCT more common for tool-in-lesion confirmation. -No evaluation against existing catheter segmentation methods and limited rationale is provided for the selection of the algorithm presented. -The manuscript makes claims about reducing procedure time by 15% but provides no evidence or explanation of how this number is reached. -The manuscript does not describe how the two spheres are placed for pose estimation after the long axis of the scope is determined. Is one sphere always at the distal end of the scope? The rationale for examining the image intensity on a grid of points is not explained and the manuscript does not explain how the method ensures that a sphere is on the distal tip. The method is not reproducible as written. Table 3 is difficult to interpret as a result. -The lesion location is required as part of the algorithm but it it not described how the nodule location is known in CBCT - is this manually identified? -The loss function experiment (table 1) contains circular logic. The authors use dice score to select the best loss function, of which combinations of dice score are an option. The objective function that contains the evaluation metric will likely appear optimal. I suggest using a different metric like mean target registration error for the tip against manual segmentation to evaluate the best objective function. -Definition of false positive is missing. What is this measuring?
- Please rate the clarity and organization of this paper
Satisfactory
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not provide sufficient information for reproducibility.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(2) Reject — should be rejected, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
-There does not seem to be a strong need for this method as applied to CBCT during bronchoscope navigation. -The method does not introduce major novelties and is not well evaluated against state of the art 3D tube-like object segmentation or pose estimation. -Details of the method are not described sufficiently that I would be able to reproduce it.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Reject
- [Post rebuttal] Please justify your final decision from above.
Although the authors may have searched the literature for work specifically related to flexible device orientation in CBCT, there are numerous prior art on device pose estimation or segmentation (which inherently solves pose estimation) in CT, ultrasound, MR, etc volumetric imaging. The presented method for CBCT does not pose a unique technical challenge; the lack of comparison against existing methods and nonstandard evaluation metrics makes the value of the proposed method difficult to understand.
The clinical impact is a weakness in my opinion. The rebuttal does not address my concerns - that CT body divergence is the main challenge for tool in lesion confirmation, and that device orientation tends to be trivial in CBCT. Tool -> pre-op CT can be challenging but I am not convinced that tool -> CBCT is the bottleneck.
Overall, the rebuttal does not address the concerns listed in my review.
Review #3
- Please describe the contribution of the paper
This paper presents a fully automated pipeline for bronchoscope pose estimation from CBCT images in robotic-assisted bronchoscopy (RAB). The approach combines a two-stage U-Net-based segmentation pipeline with a geometry-driven pose estimation method based on PCA, context-aware spherical sampling, and CAD model alignment. The method is evaluated on real-world clinical data and demonstrates sub-millimeter position accuracy, low orientation error, and strong runtime performance (~4.5 seconds per case).
- Please list the major strengths of the paper: you should highlight a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
- Strong clinical relevance addressing CT-to-body divergence in RAB.
- Use of real-world CBCT data.
- The pose estimation pipeline is particularly well-designed and represents a meaningful contribution: the integration of PCA-based axis estimation, context-aware spherical sampling for direction disambiguation, and CAD-based geometric alignment forms a coherent and effective solution tailored to CBCT imaging.
- The approach successfully combines learning-based segmentation with model-based geometric reasoning, which is both elegant and practically effective.
- The two-stage U-net segmentation, while not novel, is well-justified and supported by a reasonably large learning dataset (225 cases), sufficient for a proof-of-concept.
- Significant runtime improvement.
- Please list the major weaknesses of the paper. Please provide details: for instance, if you state that a formulation, way of using data, demonstration of clinical feasibility, or application is not novel, then you must provide specific references to prior work.
Lack of implementation and reproducibility details: The reported runtime performance is impressive given the complexity of the pipeline. However, the paper provides very limited information regarding the implementation.
Even if proprietary constraints apply, it would be highly valuable to include high-level details about the software stack, such as the programming language, main libraries/frameworks, inference backend, and general system integration approach. Such information is standard in the field and does not compromise intellectual property.
Given that achieving sub-5-second processing on full 3D CBCT volumes is non-trivial, providing these details would significantly improve reproducibility and help the community better assess the practical feasibility and engineering contribution of the work.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission does not provide sufficient information for reproducibility.
- Based on your review and your understanding of the MICCAI Scientific Code of Ethics, do you believe this submission may involve a potential ethics concern or violation?
N/A
- Optional: If you have any additional comments to share with the authors, please provide them here. Please also refer to our Reviewer’s guide on what makes a good review and pay specific attention to the different assessment criteria for the different paper categories: https://conferences.miccai.org/2026/en/REVIEWER-GUIDELINES.html
N/A
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making.
(5) Accept — should be accepted, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Overall, this paper presents a well-engineered and practically valuable contribution to CBCT-guided robotic bronchoscopy. While the segmentation component is based on established methods, the pose estimation pipeline stands out as a thoughtful and effective integration of geometric and image-based reasoning. The use of real clinical data and the strong runtime performance further strengthen the impact of the work.
The paper would benefit from additional implementation transparency and extended validation, but these do not detract from the core contribution. The work is convincing as a proof-of-concept and has clear potential for real-world clinical impact.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
Accept
- [Post rebuttal] Please justify your final decision from above.
Overall, despite some limitations in validation depth and methodological novelty, I believe the paper presents a practically valuable and technically sound contribution with clear clinical relevance, and I support acceptance.
Author Feedback
We appreciate the reviewers’ thoughtful feedback and constructive comments. Below we provide a detailed response to major concerns and will update the revised manuscript accordingly.
[R1, R2, MR: Clinical Impact] In latest RAB platforms, CBCT is used for both tool-in-lesion confirmation and navigation updates. CBCT-derived information (scope pose, lesion location) can be used to register the tracking system with CBCT coordinate frame to compensate for CT-body divergence and improve navigation accuracy.
We focus on pose estimation as a critical yet currently time-consuming step that is manual or semi-automatic, introducing latency and human error. Although this work targets the bottleneck itself rather than downstream clinical endpoints, improved pose accuracy directly enhances CBCT-navigation registration and is expected to reduce navigation error and increase biopsy yield. We agree that evaluating downstream metrics (e.g., target registration error and tool-in-lesion rate) would further strengthen the clinical validation, and it is planned for future studies.
The model was retrained and evaluated on expanded CBCT datasets from multi-vendors (GE, Siemens, Philips) in past few months with same accuracy level and no failures observed on human cases. We will continue monitoring clinical deployment and assess limitations in future work. We also have a backup option for manual pose correction intraoperatively.
[R1, R2: Existing Methods] We did another thorough literature review and could not identify a directly comparable method for pose estimation of catheter/scope-like devices in 3D volumes. Most studies focus on entire scope segmentation or endpoint localization, which differs fundamentally from our tip segmentation to address the pose estimation problem. The closest study relies on external EM data in X-ray (10.1016/j.cmpb.2022.107036).
[R1, R2, MR: Novelty] We proposed a lightweight yet effective and accurate pipeline, bridging segmentation to pose estimation in human CBCT data, especially the direction disambiguation part. Our goal is not to make every component novel, but to design a solution that will not fail in every intermediate step even in challenging scenarios for clinical use (Fig.4). The proposed pipeline is device-agnostic and can be extended to user-specific catheter-like devices across both 2D and 3D imaging modalities.
The current user-required pose estimation step takes ~90 secs while the proposed takes only ~4.5 secs, reducing total navigation time by 15% from 9.8 min to 8.3 min (reference in Sec 4.3).
[R3: Implementation] To achieve the reported runtime, the entire pipeline, as one module of the RAB navigation system, was optimized in C++. The models were trained in PyTorch, exported to ONNX format, and deployed using NVIDIA TensorRT for accelerated batch inference. For pose estimation and related image processing, we rely on VTK library, while OpenMP is used to parallelize computationally intensive steps wherever applicable.
We will include additional info about the CAD model geometry for reference as the sampling parameters for pose estimation are subject to change depending on user-specific CAD models.
[R2: Method Details] The two sampling spheres are positioned along the principal axis using CAD-derived offsets from the scope tip center, ensuring one sphere is consistently located at the distal end (Sec 2.2 and Fig.2). The nodule location is obtained either manually or automatically based on user preference (Fig.1).
[R2: Table 1] False positive refers to the number of incorrectly classified voxels, measuring over-segmentation areas due to metal implants and artifacts. The concern about circular logic would apply if the same Dice value were used both as the target being directly optimized and as the only basis for evaluating the identical data. Different compound loss functions influence training whereas the reported Dice score assesses generalization on test data, which is a standard practice
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- Please justify your decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. In case of an invitation for rebuttal, clarify which points are important to address in the rebuttal.
The paper addresses a clinically relevant problem about pose estimation for bronchoscopy using CBCT. The paper performed validation on real human data. The integration is solid. However, the technical contribution, particularly the two-stage segmentation, is relatively weak. The authors should clarify their contributions in the rebuttal and include analysis aligned with clinical needs. Failure case analysis would also strengthen this application-oriented study. Further comments can be found in the reviewers’ feedback.
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation.
Two reviewers support the acceptance, and one maintains some concern due to the contribution. I believe the paper has done a good job on the system integration, although individual parts are not new. So I lean towards accepting it due to its practical integration and validation.
Meta-review #2
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation.
I believe the work should be accepted. The primary contribution lies in the clinical evaluation and the demonstration of a functioning integrated system rather than in the novelty of the individual technical components. While I agree that more sophisticated segmentation methods could have been evaluated independently, this can reasonably be considered as future work and does not diminish the overall contribution of the study.
From a CAI perspective, this is a strong and relevant contribution that aligns well with the goals of the MICCAI Conference community, and I believe it should be included in the conference program.