Paper Info Reviews Author Feedback Meta-Review Back to top
List of Papers Browse by Subject Areas Author List

Abstract

Sensorless freehand 3D ultrasound and photoacoustic reconstruction enables flexible clinical diagnosis, yet conventional visual tracking suffers from severe long-range drift induced by speckle decorrelation and acoustic clutter. To overcome these physical degradations, we propose SCC-Net, a structure-aware channel cross-attention network for stable spatiotemporal tracking. The architecture utilizes a channel decoupling module driven by global spatial statistics for adaptive noise filtering, alongside a cross-attention mechanism capturing adjacent-frame geometric topologies. These components are jointly optimized via a progressive training strategy incorporating a motion-modulated mean absolute error to prioritize highly dynamic regions. Extensive independent experiments on a large-scale ultrasound forearm dataset and a complex photoacoustic palm dataset demonstrate the exceptional generalizability of SCC-Net in suppressing cumulative errors. Notably, compared to previous baselines, the method reduces the final drift rate by 13.72% and absolute trajectory error by 14.29% on the ultrasound dataset. Simultaneously, despite the severe acoustic clutter and intricate reciprocating trajectories of the photoacoustic dataset, the network consistently achieves state-of-the-art trajectory estimation. This enables high-fidelity three-dimensional functional visualization without external hardware, demonstrating exceptional adaptability to highly noisy environments.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/0978_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)

TUS-REC dataset: https://zenodo.org/records/11178509 https://zenodo.org/records/11180795 https://zenodo.org/records/12979481

BibTex

@InProceedings{HuaYi_StructureAware_MICCAI2026,
        author = { Huang, Yi AND Gong, Wenxin AND Zhao, Yongjian AND Shen, Li AND Lan, Hengrong AND Gao, Fei},
        title = { { Structure-Aware Sensorless 3D Ultrasound and Photoacoustic Reconstruction } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 16888},
        month = {September},
        page = {pending}
}

Reviews

Review #1

  • Please describe the contribution of the paper

    The paper proposes SCC-Net, a sensorless freehand 3D reconstruction method for both ultrasound (US) and photoacoustic (PA) imaging. The core idea is to make frame-to-frame pose estimation more robust under US speckle decorrelation, probe compression, and PA clutter, which are major failure modes in tracker-free reconstruction.

    Methodologically, the paper combines three ingredients:

    1. a structure-aware channel decoupling block that reweights feature channels to suppress unstable acoustic patterns and emphasize more stable anatomical structure;
    2. a masked cross-attention module that learns correspondences between adjacent frames using global context rather than relying too strongly on local texture;
    3. a progressive training strategy with triplet-style temporal regularization and a motion-modulated MAE to focus optimization on harder, high-motion regions.

    Empirically, the paper’s main claim is that this combination reduces long-range drift and improves reconstruction quality on a public forearm US dataset and a palm PA dataset, while also demonstrating hardware-free 3D vascular reconstruction in PA.

  • 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.

    A. Important and clinically relevant problem Sensorless freehand 3D reconstruction is practically important because removing external trackers improves workflow simplicity, cost, and compatibility. That is a meaningful direction for both US and PA, especially for handheld scanning and procedural settings.

    B. Good problem framing around modality-specific failure modes A real strength is that the paper is not just “another motion estimator,” but explicitly frames the challenge around physical degradation mechanisms: decorrelating US speckle, cluttered PA patterns, out-of-plane motion, and probe compression. That framing is well aligned with the real problem and makes the method design more coherent.

    C. Cross-modality evaluation is valuable Testing on both US and PA is stronger than a single-modality paper. The US dataset is reasonably large (50 subjects / 1200 scan sequences per the paper), and the PA experiment is valuable because PA tracking is harder and less explored than standard B-mode US. Even if the architectural novelty is somewhat incremental, showing one method that transfers across these two domains is useful.

    D. The paper appears to improve long-range stability, not just local pose The reported gains are not only on local relative errors but also on more reconstruction-relevant metrics such as ATE, final drift, and FDR. That is the right direction for this problem, because in tracker-free 3D reconstruction the real bottleneck is accumulated drift over long trajectories.

    E. Practical significance of hardware-free PA reconstruction The strongest application-level point is the demonstration of purely image-based 3D PA reconstruction of palm vasculature without external tracking hardware. Prior literature has already extended freehand reconstruction to PA/PAUS settings, especially MoGLo-Net, but this remains a valuable demonstration because PA tracking is particularly difficult due to sparse vessels and clutter.

  • 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.

    A. The technical novelty looks limited / highly compositional The paper’s main modules appear to be built from already familiar ideas:

    • the channel decoupling block is very close in spirit to SE-style channel attention, i. e. global pooling plus learned channel reweighting;
    • the masked cross-attention idea is explicitly inspired by MAE and CroCo/cross-view completion;
    • the motion-modulated loss resembles prior motion-weighted training losses already used in tracker-free US reconstruction, e. g. RecON. So the paper may be better described as a domain-tailored combination of existing mechanisms rather than a strongly new formulation. B. Novelty relative to prior tracker-free US literature is not fully convincing: The paper positions itself against “coarse-grained macro features,” but several recent works already addressed long-range dependency, sequence modeling, motion-aware learning, or stronger feature interactions in tracker-free US reconstruction. Examples include LongTerm for long-range dependency modeling, physics-guided deep learning for sensorless end-to-end reconstruction, and RecON for online/self-supervised refinement with motion-weighted losses. Because of this, the paper’s advance feels more incremental than foundational. C. The “PA novelty” claim should be stated more carefully The paper suggests novelty around applying 3D reconstruction to PA, prior study such as MoGLo-Net already studied freehand 3D photoacoustic and ultrasound reconstruction and explicitly reported PA/vasculature visualization. That means the claim should be narrowed to something like pure-PA tracking without auxiliary modalities/hardware, if that is what the authors truly mean. As written, the novelty statement risks overstating the case. D. Missing or insufficient ablation evidence This is the biggest experimental weakness for me. A paper with several components should show:
    • SACD on/off
    • masking on/off
    • cross-attention on/off
    • MMAE vs MAE
    • bidirectional/self-supervised training on/off
    • maybe US-only training vs cross-domain generalization Without these, it is hard to know which component actually matters and whether the gain comes mostly from architecture scale or training details. E. Limited detail on fairness / reproducibility The paper says comparisons are under “identical experimental settings,” but it is not fully clear whether all baselines were retrained with matched data augmentation, sequence length, curriculum, and hyperparameters. For a benchmark-heavy paper, that matters.
  • 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

    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 see this as a promising but not yet fully convincing MICCAI paper for following reasons: (1) the problem is important; (2) the method is plausible and reasonably well motivated by imaging physics; (3) evaluation on both US and PA is valuable; (4) the reported drift improvements look meaningful.

    I’d like to propose “weak reject” because: (1) the core technical ideas appear incremental and largely assembled from known ingredients (SE-like channel attention, masked modeling / cross-view completion, motion-weighted loss); (2) the novelty over recent tracker-free reconstruction work is not sharply established; (3) the paper lacks the ablation depth needed to prove that the proposed components are necessary; (4) some claims around PA novelty seem overstated given MoGLo-Net and related PAUS reconstruction work. So my summary would be: useful application paper, credible empirical improvement, but insufficiently novel and insufficiently dissected experimentally for a stronger recommendation.

  • Reviewer confidence

    Somewhat confident (2)

  • [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.

    The authors addressed well to all comments



Review #2

  • Please describe the contribution of the paper

    Structure-Aware Channel Decoupling (SACD) Module​ Adaptively filters unstable acoustic patterns and modality-specific noise using global spatial statistics. Prioritizes robust anatomical features to reduce motion-induced artifacts. Cross-Attention Mechanism for Global Geometric Topology​ Captures global spatial correspondences across adjacent frames, overcoming reliance on coarse-grained features and rigid assumptions. Enhances feature generalization by leveraging spatio-temporal context. Progressive Joint Optimization Strategy​ Combines temporal triplet loss (for smooth, physically aware feature manifolds) and a Motion-Modulated Mean Absolute Error (MMAE) loss (to focus on highly dynamic regions). Improves convergence and robustness in long-range reconstruction. Demonstration of Pure PA Domain 3D Reconstruction​ Achieves high-fidelity 3D visualization of large-scale vascular networks without external tracking hardware, showing strong adaptability to noisy environments. State-of-the-Art Performance​ Outperforms existing methods on both US and PA datasets, reducing final drift rate by 13.72% and absolute trajectory error by 14.29% on the US dataset, and achieving superior results on the challenging PA dataset. These contributions collectively advance sensorless 3D reconstruction by addressing physical degradations and enabling accurate, hardware-free clinical imaging.

  • 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.

    Structure-Aware Channel Decoupling (SACD) module: This component uses global spatial statistics to assess the temporal stability of feature channels, adaptively down-weighting those dominated by unstable acoustic patterns (e.g., US speckle decorrelation, PA clutter) and up-weighting channels encoding stable anatomical structures. This is novel because it explicitly decouples physical noise from useful structural information at the feature level, rather than relying on implicit learning alone. Cross-Attention mechanism with masked feature modeling: Unlike prior works that may overfit to local textures, the authors employ a low-ratio masked feature strategy in the latent space, forcing the network to recover missing information via global spatio-temporal context. This encourages robust, generalizable spatial correspondences across frames, which is especially important in the presence of motion and noise. The combination of these elements forms a new, principled approach to handling the specific physical degradations of freehand US/PA imaging. 2.Original Use of Data and Problem Setting​ The work is the first to demonstrate high-fidelity sensorless 3D reconstruction entirely in the pure photoacoustic domain, a much more challenging setting than ultrasound due to severe acoustic clutter and sparse vascular cross-sectional features. The authors also curate and utilize two large-scale, clinically realistic datasets: A forearm US dataset​ with diverse trajectories (straight, C-shape, S-shape) and scanning directions, including forward/backward and parallel/perpendicular motions. A palm PA dataset​ with reciprocating scanning paths, requiring stitching of multiple short trajectories. The use of these varied, unconstrained real-world sequences tests the method’s generalization far beyond controlled lab conditions. 3.Strong and Comprehensive Evaluation​ The evaluation includes multiple complementary metrics: relative translation error (RTE), relative rotation error (RRE), absolute trajectory error (ATE), final drift rate (FDR), and—for the complex PA data—the Fréchet distance (F-Dist) to assess curve similarity independent of speed. Results show consistent, statistically significant improvements: 13.72% reduction in final drift rate​ and 14.29% reduction in ATE​ on the US dataset compared to the best baseline. State-of-the-art performance on the PA dataset despite its inherent challenges. The analysis includes both quantitative tables​ and qualitative trajectory visualizations​ (Fig. 2), clearly illustrating how SCC-Net suppresses long-range drift in complex motion scenarios. 4.Clinical Feasibility Demonstration​ By achieving accurate 3D reconstruction without any external tracking hardware, the method offers a cost-effective, flexible solution compatible with a wide range of clinical environments. The ability to reconstruct large-scale vascular networks in PA without trackers highlights the potential for functional 3D imaging in interventional or point-of-care settings, where external sensors are impractical. 5.Innovative Training Strategy​ The progressive joint optimization​ approach combines: A temporal triplet loss​ to enforce smooth, physically plausible feature evolution. A Motion-Modulated Mean Absolute Error (MMAE) loss​ that dynamically emphasizes high-motion regions during training, improving accuracy where errors are most likely to accumulate. This two-stage training, along with bidirectional sequence self-supervision, mitigates directional bias and enhances long-range stability, which is often a weak point in prior sensorless methods.

  • 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.Limited Innovation in Core Architectural Components​ The structure-aware channel decoupling (SACD) module​ builds directly on established channel attention mechanisms​ (e.g., Squeeze-and-Excitation Networks [10] and ECA-Net [23]), which are widely used in computer vision to recalibrate feature maps via global average/max pooling and learned gating. While the paper adapts this idea to filter unstable acoustic patterns, the novelty lies primarily in the application domain(US/PA imaging) rather than the fundamental formulation. Similarly, the cross-attention mechanism with masked feature modeling​ draws heavily from recent self-supervised learning paradigms, such as masked autoencoders [9] and cross-view feature completion [24]. The paper’s contribution is applying these ideas to spatio-temporal tracking in medical imaging, but the underlying methods are not new. Reference: [9] He et al., “Masked autoencoders are scalable vision learners” (CVPR 2022) [10] Hu et al., “Squeeze-and-excitation networks” (CVPR 2018) [23] Wang et al., “ECA-net: Efficient channel attention” (CVPR 2020) [24] Weinzaepfel et al., “Croco: Self-supervised pre-training for 3D vision” (NeurIPS 2022)

    2.Incremental Improvement Over Recent Baselines​ The performance gains over prior methods (e.g., 13.72% reduction in final drift rate on the US dataset) are modest​ given the complexity of the problem. For example, the absolute trajectory error (ATE) of SCC-Net (21.53 mm) is only ~3.6 mm lower than MoGLo-Net (25.12 mm) on the US dataset, which may not translate to clinically significant differences in many applications. The paper does not compare against the most recent state-of-the-art methods (e.g., [26] Yan et al., 2024, which integrates multi-modal data and fine-grained context) in all experiments, limiting the comprehensiveness of the evaluation. Reference: [11] Lee et al., “Enhancing free-hand 3D photoacoustic and ultrasound reconstruction” (IEEE TMI 2025) [26] Yan et al., “Fine-grained context and multi-modal alignment” (MICCAI 2024)

  • 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 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

    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?

    Novelty: SCC-Net’s structure-aware decoupling and cross-attention design directly tackle modality-specific physical degradations. Originality in application: First successful pure PA 3D sensorless reconstruction. Data diversity & realism: Large, challenging datasets with complex motion patterns. Evaluation rigor: Multiple error metrics plus visual evidence of reduced drift. Clinical relevance: Hardware-free operation with demonstrated high-fidelity results. Training innovation: Advanced loss functions and self-supervision for improved robustness. These combined strengths make the work a significant step forward in sensorless 3D medical imaging.

  • 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 paper proposed a structure-aware channel cross-attention network, including a structure-aware channel decoupling module, a cross-attention mechanism, and a self-supervised learning strategy, to improve the pose estimation performance. Extensive experiments are conducted on two datasets, with comparisons against five 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.

    The manuscript proposes several modules and a self-supervised learning strategy for improved pose prediction.

    The experiments are extensive, involving two datasets and comparisons with multiple methods. And the results are promising.

  • 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 description of the method is not sufficiently clear, with some symbols not properly defined and certain statements lacking appropriate references.

    The test set should not be augmented.

    Although the authors compare several methods, since the dataset originates from a challenge, it would be more appropriate to compare against the top-ranking teams rather than the fourth- and fifth-ranked ones. In addition, statistical significance testing should be performed.

  • 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

    1.“If the network uniformly weighs all extracted features when estimating the displacement θi, it becomes susceptible to these modality-specific artifacts.” - Could you provide a justification for this? 2.Section 2.1: H and W should be defined when they are first introduced. And also D. 3.Eq. (3) is not clear. What does “anchor” refer to, and from which layer are the “latent features” derived? 4.In Eq. (4), the prediction is denoted as \hat{y}, while the displacement parameters are defined as \theta at the beginning of Section 2.Do these two refer to the same thing? 5.Section 2.3: What do “the first stage” and “the second stage” refer to? How is the model trained across these stages, and how are the two losses applied during training? 6.“Due to the reciprocating nature of the scanning process, data for each subject is partitioned into 13 to 14 trajectories for training, and these trajectories are stitched based on their absolute positions during reconstruction.” - This is not clearly described; it would be helpful to include more details on how the data were acquired. 7.The input sequence length is set to 5 frames to ensure baseline performance. - Why were five frames chosen? Have you experimented with other numbers of input frames? Since the method is correlation-based, it seems that using more input frames might further improve performance. 8.“Each test scan is expanded to 10 sequences to simulate complex real-world scenarios” - Applying data augmentation to the training set is reasonable; however, it is generally not appropriate for the test set. Doing so may alter the original data distribution and can lead to biased or non-comparable results. 9.In Section Quantitative and Qualitative Analysis: How is the trajectory length calculated? It would also be helpful to provide clear definitions of the other metrics, including relative translation error (RTE), relative rotation error (RRE), absolute trajectory error (ATE). 10.Could the authors also provide p-values in Table 1 to support the statistical significance of the results? 11.The performance of the proposed method is promising; however, the comparisons are made only against the fourth- and fifth-ranked teams . Why were the top three teams not included in the comparison?

  • 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?

    The method is not well described, with several terms lacking clear explanations. Although the results are promising, they lack statistical testing to demonstrate their validity. In addition, the test set should not be augmented, and the performance comparison on this dataset should include top-performing methods from the challenge.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [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.

    The authors provide explanations for the concerns raised by the reviewers. However, I remain unconvinced by their justification for not comparing with the top-ranking methods.

    Although the experimental evaluation is extensive, involving two datasets and multiple comparison methods, the manuscript is not yet publication-ready. Several sections remain difficult to follow, in addition to some of the concerns raised by the other reviewers, particularly regarding the validity and contribution of each proposed module.

    I would encourage the authors to address the reviewers’ comments carefully and consider resubmitting the work in the future.



Review #4

  • Please describe the contribution of the paper

    The paper proposes SCC-Net for sensorless freehand 3D reconstruction from ultrasound and photoacoustic image sequences. The method combines a structure-aware channel decoupling module, masked feature modeling with cross-attention, and a progressive training objective. Experiments on a public forearm ultrasound dataset and a palm photoacoustic dataset report improved trajectory reconstruction metrics over baselines.

  • 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.The motivation is clinically relevant. Trackerless 3D reconstruction for US and PA imaging could reduce hardware complexity and make freehand acquisition more practical. 2.The paper identifies realistic challenges in US/PA reconstruction, such as speckle decorrelation in US and clutter in PA imaging.

  • 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.Although the paper proposes several novel components, it does not include an ablation study to demonstrate the contribution of each component to the overall performance improvement. The reported gains could be due to increased model capacity. A model size / parameter comparison is also missing. 2.The paper claims that in component SACD, certain channels capture unstable acoustic fingerprints, such as speckle decorrelation in US or transient clutter in PA, while others extract stable structural geometries. However, this interpretation is not sufficiently supported by evidence or references.

  • 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

    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?

    The problem is important and the proposed method is potentially useful, but the current illustration is not strong enough. In particular, the lack of ablation studies makes it difficult to judge whether the proposed modules are actually responsible for the claimed gains. In addition, some explanations of the learned representations are overclaimed without sufficient evidence or references.

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [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.

    The authors have made some effort to address my concerns. However, the response still lacks key evidence to support the main claims and component-level ablations. Since too many important details and supporting analyses are still missing, this response does not change my original assessment.



Author Feedback

We sincerely thank the reviewers and meta-reviewer for recognizing the clinical relevance of sensorless US/PA reconstruction, the cross-modality evaluation, and the promising long-range stability. We address the main concerns below:

1.Fairness and protocol clarification (R1,R3). All results in Table 1 used the same split, data budget, augmentation policy, schedule, and evaluation protocol thus the gains do not come from longer training or more sampled data. The test set was not augmented: the “10 sequences” are temporal windows with different starting frames from the original scans, not image transformations. PA images received the same fixed brightness enhancement for all methods. We will clarify that the 5-frame input is a local motion window balancing temporal context and memory cost, not an optimality claim. We will also define H/W/C/D, anchors, latent features, stages, metrics, trajectory length, PA stitching, and Eq. 4 notation, and report p-values for existing runs.

2.Component effect and model capacity (R1,R4,AC). We appreciate the reviewers’ concern about component-level evidence. Since rebuttal guidelines do not allow additional experiments or substantial experimental changes, we focus here on the current evidence and design rationale. Table 1 provides a controlled full-model comparison under the same image-only setting and matched data traversal, directly evaluating accumulated drift. The observed gains are consistent with the intended role of each design: SACD preserves structure-related cues while reducing unstable responses from ultrasound speckle and photoacoustic artifacts,Latent masking reduces dependence on local textures,Cross-attention introduces neighboring-frame context, MMAE weights larger-motion samples and bidirectional learning reduces scan-direction bias. The gain is unlikely to come from model size, since our model uses a lightweight ConvLSTM and has 24.47M parameters, compared with 27.91M for MoGLo. Our variants are also close in size, from 24.30M to 24.47M. We will report model capacity and revise the module descriptions so that the claims remain aligned with the reported evidence.

3.Related work and comparison scope (R1,R2,R3,AC). We will better position our work with LongTerm, RecON, physics-guided reconstruction, and Yan et al. (FiMA). We do not claim a leaderboard ranking over tracker-free US literature, since methods differ in input modality, supervision, online refinement, or protocol. Table 1 instead provides a matched image-only comparison under the same split, data traversal, and training budget. FiMA uses ReMamba with multiple IMUs and multimodal alignment; including IMUs changes our setting, while removing them misrepresents FiMA. Thus, we do not claim superiority over FiMA. TUS-REC top entries use ensembles, pretraining, checkpoint averaging, and scan-direction/offset fusion; they are useful references, not direct competitors.

4.Novelty, PA claim, and clinical interpretation (R1,R2,R4,AC). We will make the novelty statement more precise. Some blocks relate to channel attention, masked feature learning, and bidirectional modeling. Our contribution is the task-specific organization of these designs for US/PA reconstruction under acoustic artifacts, local texture instability, and scan-direction bias. SACD will be described as motivated by ultrasound speckle and PA artifacts, not as proof of physical factor separation. The masking design is lightly inspired by CroCo and is not claimed as a new masking paradigm. Since MoGLo-Net has studied freehand PA/US reconstruction, our PA setting is pure image-based reconstruction without external tracking hardware or US structural priors. The ATE gap should be read with FDR and absolute errors under the same sampled data traversal and training budget, not as a standalone clinical margin. Our main claim is reduced drift and improved long-range stability in the image-only setting.



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 the problem of sensorless freehand 3D reconstruction in ultrasound and photoacoustic imaging. The three reviewers acknowledge the relevance of the task and recognize some novelty in the proposed combination of blocks. The ability to perform well across modalities (ultrasound and photoacoustics) was welcomed, especially given the difficulty of the photoacoustic case. Experiments were conducted on several curated datasets (forearm, palm). Results reported across several metrics show improved long-range stability of pose estimation. The main drawbacks raised were the incremental novelty, the unclear improvement over the state-of-the-art tracker-free US literature, and a limited ablation study. Claims of novelty in photoacoustics should be tempered by acknowledging prior work.

  • After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.

    Reject

  • Please justify your recommendation.

    The paper remains borderline after the rebuttal, with two acceptance and two rejection votes. The novelty is compositional but acknowledged, and generalization on two modalities is interesting. Several clarifications and indications to moderate certain claims have been addressed in the rebuttal. The rebuttal also explains how some of the related work being compared to had access to additional data or employed more complex post-processing strategies, which seems reasonable. The remaining remarks for R3 and R4 are the lack of an ablation study, and R3 also recalls a lack of clarity in their after-rebuttal comments.



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.

    The authors have provided convincing justifications to the concerns raised by the reviewers.



Meta-review #3

  • 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.

    The author’s response addresses the reviewers’ concerns.



back to top