Abstract

Visceral Adipose Tissue (VAT) is a key biomarker for cardiometabolic disease risk, but its accurate quantification typically requires Computed Tomography (CT) or Magnetic Resonance Imaging (MRI), which are costly and not routinely available for large-scale or longitudinal assessment. Dual-energy X-ray Absorptiometry (DXA), on the other hand, offers a low-radiation, cost-effective alternative. However, existing research on VAT estimation from DXA remains limited and relies on single-view–based methods. In this work, we introduce ‘VisiFAT’, a multi-view, multi-modal framework for estimating VAT from lateral and posteroanterior spine DXA scans, together with patient demographic data including sex, age, height, and weight. The proposed approach employs a dual-branch convolutional architecture coupled with a cross-view attention module that enables adaptive feature exchange between views. By jointly modelling complementary DXA projections and patient-level information, the framework captures three-dimensional body composition cues that are not observable from either view alone. Evaluated on a large cohort, the proposed approach outperforms single-view and conventional fusion baselines, achieving lower prediction error and stronger correlation with reference VAT measures. To assess clinical relevance, we further examined cross-sectional associations between both estimated and reference VAT and the prevalence of atherosclerotic cardiovascular disease and diabetes, demonstrating comparable risk stratification performance. These results support ‘VisiFAT’ as a scalable tool for VAT estimation and cardiometabolic risk assessment.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/0828_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/arroobamaqsood/VisiFAT-Estimating-Visceral-Fat-from-Dual-View-DXA-scans-using-Cross-View-Attention

Link to the Dataset(s)

N/A

BibTex

@InProceedings{MaqAro_VisiFAT_MICCAI2026,
        author = { Maqsood, Arooba AND Tariq, Abdullah AND Mundt, Marion AND Sim, Marc AND Suter, David AND Harvey, Nicholas C. AND Leslie, William D. AND Schousboe, John T. AND Lewis, Joshua R. AND Gilani, Syed Zulqarnain},
        title = { { VisiFAT: Estimating Visceral Fat from Dual-View DXA Scans Using Cross-View Attention } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 16896},
        month = {September},
        page = {pending}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    The paper uses a multi view (lateral and posteroanterior spine) DXA, demographic, data, and dual-branch convolutional approach with a cross-view attention module estimation of VAT. Prediction target is VAT mass from whole body DXA scan. The whole-body DXA can estimate VAT but is not routinely performed, while spine DXA scans are already collected in patients coming for osteoporosis screening.

  • 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 approach, while technically not novel, provides an interesting application of the technique to solve a practical medical 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 related work section seems a bit thin. Using cross-view attention with consistency loss is a standard architectural adaptation rather than a major novel approach. In the VAT estimation literature, the comparisons lean heavily on a key prior model. It will be helpful if the table title uses RMSE (g) and MAE (g) with units. Additionally, Table 1 and Table 3 show the proposed model’s sMAPE at 17.85%, but Table 2 cites Cross-view Attn. + CVC regularisation at 17.25%. Are these referring to the same final model? The paper state that the framework captures three-dimensional body composition cues not observable from either view alone. Is there any direct proof? The comparison to Maqsood et al. employs LS with age, weight, and height, while the proposed model uses LS+PAS with sex, age, weight, and height, which isn’t a controlled apples-to-apples comparison. For a better setting, they could also use the same backbone, same optimization recipe, possibly a “proposed architecture with LS only” control.

  • 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?

    The approach is a novel application of a known technique to serve a clinical need. However the background relevant work needs to be strengthened as is a better comparison to existing approaches.

  • 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 VisiFAT, a deep learning framework for estimating visceral adipose tissue (VAT) from standard osteoporosis DXA scans. Building upon previous work by Masdooq et al., which focused solely on lateral spine (LS) imaging, this study incorporates posteroanterior (PAS) views to capture a more comprehensive representation of the abdominal anatomy, simulating a volumetric view. The technical innovation of the paper is a dual-view architecture that uses a cross-attention mechanism for feature fusion, further guided by a cross-view constraint (CVC) to ensure anatomical consistency across views. Beyond technical performance, the authors validate the clinical utility of their estimated VAT by evaluating its association with diabetes and atherosclerotic cardiovascular disease (ASCVD) and comparing these results with ground-truth VAT measurements.

  • 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 paper presents a detailed model architecture featuring a novel cross-view attention fusion (CVAF) mechanism designed to integrate different modalities effectively. To maintain consistency across perspectives, the authors implement CVC regularization, which ensures that VAT estimations remain stable regardless of the input image view. The experimental section is rigorous, offering extensive benchmark comparisons and ablation studies that clarify the specific contributions of each technical novelty and demonstrate how the approach outperforms conventional multi-view fusion strategies. Most importantly, the authors ground their technical findings in real-world impact by including a clinical validation that correlates their results with diabetes and cardiovascular events.

  • 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 authors demonstrate clear improvements with the VisiFAT model compared to single-view and non-attention baselines. To help strengthen the impact of the work, there are a few areas where further clarification would be very beneficial: Regarding the choice of ground truth, could the authors clarify the decision to use CoreScan estimations? Since this study utilizes the UK Biobank, do we have access to MRI data for many participants to see how VisiFAT performs against MRI-derived VAT? This would be particularly helpful, since some studies show that DXA-based estimates, such as CoreScan, can exhibit systematic biases at the extremes of the VAT distribution. There are also a few questions regarding the preprocessing steps that might help readers better understand the methodology. Given the broad size ranges for the LS and PAS views, how was anatomical consistency maintained across scans that may cover different regions of the body? If the views are focused specifically on the spine rather than the whole body, it would be helpful to clarify what features the model is primarily learning from. Additionally, it would be useful to know how the authors maintained a constant physical resolution during the resizing process, and what the reasoning was behind the specific cropping and padding strategy used. In the multivariable-adjusted logistic regression models, what was the motivation for including the scan year as a covariate? Understanding how this variable interacts with the model’s predictions would be very helpful for evaluating its generalizability to other clinical settings. Regarding the validation strategy, while the 80/10/10 split is a rigorous way to use the available data, knowing if the model was tested on a completely isolated dataset would add confidence in its real-world performance.

  • 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

    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?

    The manuscript is overall very well written and organized. The authors present thorough experiments and ablation studies to highlight the contribution of each component in the methodology.

  • 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

    A novel multi-view, multi-modal deep learning framework (VisiFAT) that leverages cross-view attention to estimate VAT from DXA scans, improving accuracy over single-view methods while maintaining clinical relevance for disease risk assessment.

  • 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 paper addresses an important clinical problem, namely scalable VAT estimation without relying on costly imaging modalities such as MRI or CT.
    • The use of multi-view DXA data is well motivated and represents a meaningful extension beyond existing single-view approaches.
    • The integration of demographic information alongside imaging data is appropriate and practically relevant.
    • The clinical analysis, particularly the association with cardiovascular disease prevalence, is a strong aspect of the work and enhances its translational value.
    • The method shows quantitative improvements over baselines, suggesting the benefit of the proposed multi-view fusion strategy.
  • 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.

    Conceptual and Positioning Issues

    • The claim of extracting 3D information from only two DXA views appears somewhat strong and may not be fully justified. This statement should be softened or better supported.
    • The Introduction contains overly strong claims regarding MRI. There are large-scale longitudinal studies that successfully use MRI for SAT and VAT quantification. Moreover, MRI provides additional advantages such as fat fraction estimation, which is not available with CT or DXA. While the argument about cost and efficiency is valid, the current wording should be revised to avoid being misleading. Emphasizing the accessibility and low cost of DXA would be a more balanced positioning.

    Methodological Clarity

    • The description of the dual-view image encoders is unclear. Specifically, it is not explained where the number “3” comes from (e.g., whether it corresponds to three slices per case).
    • When flattening feature maps into token sequences, it is unclear how spatial information is preserved in the latent representation.
    • The cross-view attention fusion (CVAF) module is insufficiently distinguished from a standard transformer. The current description largely mirrors standard attention mechanisms, making the novelty unclear.
    • The cross-view consistency regularization requires further clarification. The use of a squared distance loss is not well justified; in contrast, contrastive losses may provide a more flexible and softer alignment mechanism. A clear formulation and rationale are needed.

    Baselines and Experimental Design

    • The multi-view baseline is not described in sufficient detail. More explanation is needed to allow proper comparison with the proposed method.
    • The single-view approach is referenced in the Results but not adequately described in the Methods, making it difficult to interpret the comparison.

    Paper Organization

    • The Results and Discussion should be a separate section, rather than as a subsection of the Methods.
  • 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?

    The paper presents a clinically relevant and practically motivated approach with promising results. However, several claims are overstated, and important methodological details are unclear or insufficiently justified. Addressing these issues, particularly improving clarity, strengthening baselines, and moderating claim could improve the work.

  • Reviewer confidence

    Very confident (4)

  • [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

Thank you, all the reviewers, for taking the time to carefully review our paper and provide valuable feedback. We appreciate the insightful comments and will clarify these points in the camera-ready version of the paper.




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.

    The reviews are consistently positive, with three weak-accept recommendations. The paper addresses a clinically relevant problem by estimating visceral adipose tissue from routinely acquired dual-view spine DXA scans, and the reviewers recognize the practical value of combining LS/PAS views, demographic information, cross-view attention, ablation studies, and clinical association analyses with ASCVD and diabetes.

    The main concerns are moderate methodological novelty, some overstatement of the “3D cues” claim, and the need for clearer baseline comparisons and preprocessing details. These issues appear addressable in revision and do not substantially undermine the contribution.



back to top