List of Papers Browse by Subject Areas Author List
Abstract
Accurate estimation of left ventricular (LV) orientation is essential for cardiac magnetic resonance (CMR) imaging and downstream analysis. Existing methods typically formulate orientation recognition as discrete view classification or rely on multi-slice geometric intersection, limiting their ability to model continuous 3D orientation and generalize across arbitrary slices. This work introduces a novel paradigm: Joint LV localization and 3D orientation estimation from a single CMR slice. To investigate this setting, representative orientation-aware detection frameworks are adapted to the CMR domain, and their limitations are analyzed. Upon that, we propose the Polar-Coupled Circular (PCC) embedding that provides a continuous and unambiguous orientation representation to address the limitations. Meanwhile, a scalable benchmark is constructed through automatic slice sampling from volumetric CMR segmentation datasets. Extensive experiments on four datasets demonstrate strong performance, achieving an average mIoU of 86.18% and an average angle deviation of 3.39°. This study establishes a new task setting for single-slice LV orientation modeling and provides a geometry-consistent framework for spatially informed CMR analysis. Code is available at https://github.com/yuyi1005/cmr-3d-ood.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/0016_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/yuyi1005/cmr-3d-ood
Link to the Dataset(s)
N/A
BibTex
@InProceedings{YuYi_Automatic_MICCAI2026,
author = { Yu, Yi AND Liu, Yixuan AND Zhang, Ziyu AND Martin, Parker AND Bu, Zhenyu AND Han, Yuchi AND Xue, Yuan},
title = { { Automatic LV Localization and Short-Axis Plane Estimation from Arbitrary CMR Slice } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16895},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
This work presents a novel task setting in the context of cardiac magnetic resonance (CMR) imaging, namely the joint estimation of left ventricle (LV) location and 3D orientation (as defined by the authors) from a single CMR slice. The authors review representative state-of-the-art (SOTA) orientation detection frameworks from other domains and discuss their limitations. They then propose a new methodology introducing a continuous and geometrically consistent representation of 3D orientations, which is shown to outperform the studied SOTA methods in this new task setting across four datasets. Finally, they demonstrate its added value in CMR plane prescription (a more conventional task), again outperforming task-specific SOTA methods.
- 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.
Task novelty and benchmarking: The study introduces an interesting new formulation for LV orientation recognition (as defined in the manuscript) in CMR. Rather than relying on view classification as a proxy or predicting short-axis (SAx) stack orientation via multi-view geometric aggregation (typical of plane prescription), the authors directly estimate a 3D vector representing LV orientation (i. e. SAx stack normal) from a single arbitrary slice. This formulation is applicable to both settings and may further enable new automated pipelines, particularly in single-view scenarios, potentially stimulating further research. The authors also report baseline performance for four object detection methods and claim to release the data curation pipeline used to benchmark this task across four public datasets.
Methodological novelty: Beyond defining a new task and benchmarking prior SOTA methods, the authors introduce a novel orientation representation and associated loss function, yielding improved performance over existing strategies.
Pertinent and comprehensive experiments: The proposed method is validated in the new task setting and supported by a concise ablation study on key embedding hyperparameters. Additionally, evaluation on a traditional task (SAx plane prescription) shows improved performance over SOTA methods, supporting the generalization and versatility of the approach.
- 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.
Small validation study: Although the comparison against multi-view plane prescription methods is welcome (as noted above), it is conducted on a very limited dataset (four volunteers). Since this is the only experiment evaluating generalization in a real clinical task (as opposed to the proposed setting), validation on a larger dataset (or those used in prior plane prescription studies, if possible) would have strengthened the manuscript.
Misleading nomenclature: Although of lesser importance, some terminology (e. g. , “spherical circular harmonics”, “3D LV orientation”) is potentially misleading. Details are provided below.
- 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 authors claimed to release the source code and/or dataset upon acceptance of the submission.
- 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
1) While the motivation for single-plane 3D object orientation prediction is clear (and supported by the results), the comparison to monocular 3D detection paradigm is somewhat odd. In natural image settings, object orientation can be visually inferred from perspective cues relative to the camera. In CMR, however, images are 2D cross-sections of the body rather than perspective projections (cf. Fig. 1a). The problem thus appears ill-posed at first glance, with the hypothesis relying on the model’s ability to infer orientation purely from in-slice anatomical features. Please consider discussing this aspect when introducing the task or reframing it. 2) The term “3D LV orientation”, while not incorrect, may be somewhat misleading. Section 3.4 clearly indicates that the model is trained to predict the orientation of the SAx stack Although this serves as a proxy for LV orientation, discrepancies may arise in practice; for example, due to suboptimal view planning or anatomical variability such as pronounced LV curvature (e. g. , “banana-shaped” ventricles), whether congenital or resulting from pathological remodeling. Referring to SAx orientation or plane estimation (as in Fig. 4 and the title) thus seem more precise. 3) While the method builds on FCOS, alternative formulations such as the CenterNet family address similar detection and pose estimation tasks using center-based representations. It would be helpful to position your approach within this broader context, clarify how it compares to these methods, and justify the choice of FCOS as the baseline. 4) The term “spherical circular harmonic” is mathematically misleading. “Spherical” refers to functions defined on the sphere (S2), whereas “circular” refers to S1.Equation (1) describes a circular harmonic, later extended in eq. (2) with a z-component, enabling representation of 3D orientation but not constituting a true spherical harmonic expansion. Please consider revising this terminology. 5) The SOTA methods in Table 1 appear to have been implemented by replacing the proposed loss (and embedding) with their respective formulations. If so, this should be clearly stated. Section 4 (“All models adopt (…)”) suggests this, but clearer wording (including in Table 1) would improve clarity. 6) Please discuss the potential evaluation leakage arising from applying the same data augmentation to both training and testing data. Given the anisotropic resolution and inter-slice spacing of the original stacks, the model may exploit interpolation or slice-thickness artifacts rather than true geometric cues, potentially overestimating generalization. It is important to note, however, that the additional plane prescription experiment partially mitigates this concern.
Minor remarks:
- The phrase “the devised data construction pipeline” in the abstract is unclear at that point in time; please consider rephrasing.
- Equation 6 includes four loss terms, but Fig. 3 shows only 3.Please consider modifying the figure for completeness.
- 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?
The manuscript presents a novel and well-motivated task formulation, supported by a thoughtful analysis of related SOTA methods and a benchmarking effort across multiple datasets. The proposed orientation representation and associated loss are technically sound and demonstrate consistent performance improvements. Most of my comments concern task framing, terminology, and clarity of presentation, and can be addressed without substantial changes to the core contribution. A potential concern relates to evaluation leakage due to data augmentation in test data; however, this is partially mitigated by the additional plane prescription experiment and the inclusion of a dataset with isotropic volumes. Overall, the work offers meaningful contributions and is of interest to the community.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #2
- Please describe the contribution of the paper
This paper introduces a new task setting: joint left-ventricular localization and 3D orientation estimation from a single arbitrary CMR slice, rather than discrete view classification or multi-slice geometric intersection. The authors propose a Spherical Circular Harmonic (SCH) embedding for continuous 3D orientation regression. In addition, they construct a benchmark from volumetric CMR datasets using a stochastic slice generation pipeline and report experiments on four datasets. Overall, I view the main contribution as a combination of new task formulation, benchmark construction, and a practical orientation representation, rather than a fundamentally new detection framework.
- 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 main strengths of the paper are the interesting task formulation, the practical benchmark construction pipeline, and the clear analysis of existing method limitations. The work is strongest as a problem-formulation and benchmark paper, rather than as a highly novel method paper or a strongly clinically motivated study.
- 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 main weaknesses are the underdeveloped clinical motivation and the limited methodological novelty. The paper does not clearly establish the concrete clinical role of the proposed orientation task, and the method is largely an adaptation of an existing FCOS-style framework with a new orientation embedding.
- 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 authors claimed to release the source code and/or dataset upon acceptance of the submission.
- 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.
(3) Weak Reject — marginally below the acceptance threshold, but would not mind if accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
I am not convinced that the submission is strong enough for acceptance in its current form for three reasons: The main strength of this paper is its application-oriented task formulation rather than its methodological novelty. The proposed single-slice LV localization and 3D orientation estimation setting is interesting, and the benchmark construction pipeline is potentially useful.
However, the clinical utility is not clearly established. The paper mentions several possible applications, but does not explain clearly in which clinical workflow this task would be used or what concrete benefit it provides over existing practice.
Methodologically, the novelty is modest. The framework mainly builds on an FCOS-style detector, and the SCH embedding is better viewed as a reasonable representation refinement than a substantially new method.
Finally, the experiments rely heavily on a generated benchmark, while real-world validation is limited. Overall, the work is promising, but I find it currently slightly below the acceptance threshold.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Review #3
- Please describe the contribution of the paper
- This work is the first to formally define the task of jointly estimating the left ventricle (LV) location and its 3D orientation from a single CMR slice.
- Proposing a data generation pipeline that automatically creates arbitrary-view training samples without requiring additional annotations.
- It leverages existing segmentation datasets to synthesize diverse training views, enabling efficient dataset expansion without manual labeling.
- 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.
Similar with its contributions.
- The paper introduces a well-motivated and novel problem formulation, addressing the joint estimation of LV location and 3D orientation from a single CMR slice.
- The proposed approach is technically sound and well-designed.
- A key strength lies in the annotation-free and efficient data pipeline, which leverages existing segmentation datasets to generate diverse training samples without additional labeling effort.
- 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.
Several concerns should be addressed in the rebuttal. 1.Lack of implementation details for the 3D extension of the baseline method: The paper only states that PSC is extended to 3D by adding a z-channel, but does not clearly explain how this extension is actually implemented.
2.Lack of validation for manual annotations: The quality and reliability of the manually annotated data are not sufficiently verified.
3.Need for additional ablation studies: It is unclear whether the number of sampled slices per volume (20) is sufficient. The authors are encouraged to evaluate different sampling strategies and also test the proposed pipeline with various backbone architectures.
4.Limited sample size in multi-view clinical evaluation: The clinical comparison is conducted on only four volunteers. While it is understandable that medical data is difficult to collect, this sample size is still quite small. Validation on a larger number of patients or volunteers is recommended to support the findings and contributions of this work.
- 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 authors claimed to release the source code and/or dataset upon acceptance of the submission.
- 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.
(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?
This paper is well organized and tackles new challenges.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Author Feedback
We sincerely thank the reviewers and the Area Chair for their insightful comments on our manuscript. We are deeply honored by your recognition of our work. We would like to take this opportunity to address some key concerns. We will also carefully incorporate other comments in the preparation of final paper.
1) The monocular 3D detection analogy and task formulation (R1). We respectfully suggest that, although CMR images are cross-sectional slices rather than perspective projections, different slice orientations still produce distinct anatomical appearance patterns, as evidenced by the success of cardiac view classification. Our task can therefore be viewed as extending discrete view classification into a continuous orientation regression problem, where the model learns continuously varying anatomical patterns associated with slice orientation. We will clarify this discussion in the introduction.
2) Application-oriented but clinical utility not clearly established (R2). We respectfully argue that the proposed task has two concrete applications. First, cardiac view classification is already widely used for automated data curation in large-scale CMR model development, including filtering, categorizing, and retrieving. Our continuous orientation prediction naturally supports all such applications while additionally enabling flexibility beyond discrete view labels. Second, as shown in Fig. 5, we evaluate the method on real-world plane prescription cases and demonstrate that it is sufficiently accurate for practical clinical plane prescription. Methodologically, while the proposed framework is simple, it indeed addresses the instability issues of existing representations, which, to the best of our knowledge, have not been explicitly addressed in literature.
3) Details of 3D extension of the baseline method (R3). We extend the 2D baseline by adding an additional regression channel for the z component of the normal vector, followed by a sigmoid activation to constrain the prediction to [0,1]. We will make it clearer in the paper.
4) Quality and reliability of manual annotation (R3). For Fig. 5, the data was collected during the standard clinical workflow, where localizer images are first acquired and SAX views are subsequently prescribed manually by the operator. The annotations are obtained directly from the prescribed SAX acquisition parameters stored in the DICOM metadata during the actual clinical scan, which we believe provides reliable ground truth for evaluation.
Meta-Review
Meta-review #1
- Your recommendation
Provisional Accept
- 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.
Reviewers agreed that the paper introduces a novel and well-motivated task. Strengths include the clear problem formulation, the annotation-efficient data generation strategy, and consistent performance improvements. While some concerns were raised regarding limited clinical validation, methodological novelty, and small-scale real-world experiments, these do not undermine the overall contribution. I recommend acceptance. Please carefully address the comments of R1 in the final version.