List of Papers Browse by Subject Areas Author List
Abstract
The EOS imaging system is a low-dose, biplanar X-ray modality offering high-fidelity anatomical visualization in standing and seated positions, benefiting total hip arthroplasty (THA) by providing accurate skeletal alignment and implant positioning pre- and postoperatively. Evaluating bone mineral density (BMD) and muscle mass before surgery is useful for predicting outcomes and tailoring rehabilitation. Although CT and DXA can assess these metrics effectively, they increase cost and radiation exposure.
Recent advances in deep learning have enabled BMD and muscle mass estimation from plain radiographs, among which one promising approach with potentially high generalizability to new modality utilized 2D–3D registration with CT of the same patient in training data preparation. However, limited EOS availability constrains large data collection. We devised and validated a deep learning framework to predict BMD and muscle mass from EOS images by fine-tuning a model trained on plain radiographs. Our dataset comprised 77 pairs of pre- and postoperative EOS images and CT scans, then underwent 2D–3D registration to create paired training data.
Our contribution is two-fold: 1) we achieved reliable BMD and muscle mass estimation in THA cases with minimal training data, and 2) we experimentally demonstrated that only 40 paired EOS–CT images were sufficient to reach high accuracy, supporting feasibility in resource-limited settings.
Future work will extend this approach to broader patient populations and anatomical sites while performing external validation to assess potential domain shifts across different facilities.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/5182_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{SueKaz_Estimating_MICCAI2025,
author = { Suehara, Kazuki and Gu, Yi and Otake, Yoshito and Uemura, Keisuke and Okamoto, Masashi and Tokunaga, Kunihiko and Talbot, Hugues and Sato, Yoshinobu},
title = { { Estimating Bone Mineral Density and Muscle Mass from EOS Low Dose X-ray Imaging System } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15970},
month = {September},
}
Reviews
Review #1
- Please describe the contribution of the paper
The study proposes a method to estimate patient’s femoral bone mineral density and lower limb muscle mass based on the anterior-posterior EOS image. For this purpose, the authors use earlier DL framework trained based on QCT and DXA data to estimate BMD from X-ray images and tune and re-train the architecture for EOS images. The authors report correlations and mean absolute error between CT-derived and EOS-estimated aBMD.
- 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.
Osteoporosis, when causing fragility fractures, decreases quality of life and increases mortality and is therefore an important disease to address. Muscle loss leading to sarcopenia and frailty is also a relevant subject. EOS is a relatively new imaging modality and, from the perspective of clinical routine, still finds its optimal application and patient groups who would most benefit from these scans. Therefore, studies exploring the possibilities of the new imaging modality are welcome.
- 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 study lacks information on how the authors derived the ground truth BMD from CT. Therefore, it is difficult to evaluate the true accuracy of the method. Authors refer to [4,5] (Gu et al,Medical image analysis 2023 and Gu et al., MICCAI 2023) where they adopted the architecture, but in [4] they used DXA and Quantitative CT (QCT) images i.e. CT with BMD calibration phantom included in the image to get the ground truth vBMD and aBMD. Did authors use QCT images, or did they rely on some earlier architecture to estimate BMD? In Fig. 3a authors show that aBMD derived from pre and post operation CT correlates well, but this does not tell how correct the absolute CT-derived aBMD values are. Clearer description and more information are needed. If BMD is estimated and not directly measured, there is a risk of propagation of error arising from the chain of estimations. EOS devices are quite rare. Therefore, it is not clear how big the impact the method would give. It remained also unclear why to estimate BMD from EOS and not measure it with DXA device? Very few additional cases would be captured in opportunistic screening with EOS since most patients ending up with EOS scan have visited also in CT, X-ray or DXA within a year. Therefore, the clinical impact of the proposed method seems moderate. Best of my knowledge, DXA scanners are also cheaper than EOS, so purchase costs favor DXA for these two if a medical unit plans to buy a scanner to measure BMD and muscle mass. BMD and muscle mass estimation could benefit from using both AP and lateral images the special feature the EOS device provides. The method is now close to estimating these parameters from normal X-ray, which are already published [4,5].
- 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.
- 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/2025/en/REVIEWER-GUIDELINES.html
“By leveraging the low-dose capability of EOS imaging and integrating advanced machine learning techniques…” Has EOS lower dose than DXA?
“Because EOS imaging delivers low-dose, biplanar radiographs, it permits frequent postoperative monitoring with minimal radiation exposure.” Frequent monitoring should not be an aim, and all monitoring must have clinical justification.
Fig 3 (c) please give unit for MAE.
In Table 2 caption, please give information what is compared where so that reader can understand it without going to the text where table 2 results are described.
- 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 — could be rejected, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
In the current form, it is unclear how BMD is derived from CT. Did authors use QCT images but did not mention it, was BMD derived using earlier method in [4] or did authors use some other asynchronous or phantomless calibration technique? More information is needed.
Methodological advance in the study is moderate and clinical benefit for the technique which would derive BMD from EOS device is moderate. For these two, deriving muscle mass from EOS might have higher clinical relevance, especially if the method would use both projections EOS provides.
- 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.
Authors will add section “QCT Calibration Protocol”, which is an important addition for those who want to repeat the study. EOS is relatively new modality which has clear benefits over other ones such as imaging in weight-bearing posture with two simultaneous projections and long field of views. Studies suggesting new use cases forthis modality are are valuable for the field.
Review #2
- Please describe the contribution of the paper
This study proposes a deep learning-based pipeline for estimating bone mineral density (BMD) and muscle mass from EOS low-dose biplanar X-ray images, particularly in the context of total hip arthroplasty (THA). By adapting and fine-tuning models originally trained on conventional radiographs using a small number of EOS–CT paired data, the authors demonstrate that accurate BMD and muscle mass estimation is achievable with as few as 40 EOS-CT pairs. The pipeline incorporates 2D–3D registration, digitally reconstructed radiographs (DRRs), and deep learning models (BMD-GAN and MSKdeX). The findings indicate that fine-tuning substantially improves prediction accuracy compared to training from scratch or linear calibration.
- 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.
Clear clinical relevance: The application to THA preoperative planning and postoperative monitoring using low-dose EOS images addresses an important need for safer and more accessible musculoskeletal evaluation.
Efficient use of limited data: Demonstrating that only ~40 training cases are sufficient for stable performance is impactful, especially for modalities like EOS that are less widely available.
Technically sound methodology: The combination of DRR generation, 2D–3D registration, and transfer learning from conventional X-ray to EOS is well-justified and implemented based on prior work.
Dual task evaluation: Simultaneous prediction of BMD and muscle mass enhances the utility of the method for broader musculoskeletal assessment.
Robust evaluation: Use of PCC, ICC, MAE, and reproducibility analysis (pre/post-op, training size effects) provides a comprehensive assessment.
- 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.
Limited generalization validation: Data is collected from only two institutions, with relatively small sample size (n = 77), and generalization to other patient populations or acquisition settings remains untested.
Lack of qualitative analysis: Visual results (e.g., sample predictions, heatmaps, overlays) are not shown, making it difficult to assess spatial accuracy or failure cases.
Dependence on CT for ground truth: The reliance on CT-derived BMD/muscle mass values is necessary but may limit clinical applicability where CT is unavailable.
Model interpretability: The methods used (BMD-GAN and MSKdeX) are not fully detailed in terms of architecture and feature attribution, which may limit insight into what anatomical cues drive the predictions.
- 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.
- 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/2025/en/REVIEWER-GUIDELINES.html
Clarify whether the models (e.g., BMD-GAN, MSKdeX) and training pipelines will be made publicly available.
Include visual examples comparing predicted vs. ground truth measurements, especially for muscle volume estimation.
Justify the decision to use only frontal EOS images; could lateral or combined input improve performance?
Provide additional architectural details of the BMD and muscle mass models, including input size, augmentation, and training hyperparameters.
Consider discussing clinical threshold values or interpretability of outputs for real-world integration (e.g., diagnosing osteoporosis or sarcopenia).
- 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 — could be accepted, 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 practically motivated and technically sound approach to estimating BMD and muscle mass using a relatively underutilized but clinically relevant modality. The use of limited paired data and transfer learning demonstrates feasibility for low-resource scenarios. While the study is promising, limitations in generalizability, lack of qualitative validation, and absence of code/data availability slightly reduce its impact. Nonetheless, it presents an interesting contribution that would be valuable to the MICCAI community.
- Reviewer confidence
Somewhat confident (2)
- [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 proposes a deep learning framework to predict the bone mineral density and muscle mass from EOS images, which is useful for predicting outcomes and tailoring rehabilitation before the total hip replacement surgery.
- 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 is well-written, with logical flow and clear. The proposed framework can acheive relaible estimation of bone mineral density and muscle mass with minimal training data, which shows high clinical potential. Fully experimented on real datasets.
- 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 proposed framework appears to be a stitching together of multiple pre-existing works. The authors use [6] to obtain the musculoskeletal segmentation labels, use [9] to perform registration, finially use the frameworks [4, 5] to estimate bone density and muscle mass. So the novelty of the work is moderate.
The authors should have been more detailed and clear about what targeted improvements or modifications were made to the existing framework [4, 5] when the input is not X-ray but EOS images.
- 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.
- 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/2025/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 — could be accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Although the the novelty of the work is limited from my understanding, it still shows high potential clinical values.
- Reviewer confidence
Somewhat confident (2)
- [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 already kind of answered all main questions from the reviewers, but I am still wrong about the novelty.
Author Feedback
We appreciate the reviewers’ constructive feedback and will revise the manuscript accordingly in the final submission.
1) Validity of CT-derived ground-truth BMD (Reviewer #3) We apologize for the insufficient detail. The final version will include a dedicated subsection titled “QCT Calibration Protocol.” For the pre-training dataset (600 CT scans with embedded phantoms), we employ the phantom-based calibration pipeline validated in our prior study (Uemura et al., Arch Osteopor 2022; r = 0.95 vs DXA, SEE = 0.050 g/cm²). For the EOS fine-tuning set (77 CT scans without phantoms), we apply a fixed calibration curve derived from the same scanner and protocol. This “phantom-less” approach, as shown in the same study, retains high correlation with DXA (r > 0.94) under protocol-matched conditions. The revised manuscript explicitly lists (i) whether a phantom was present, (ii) the calibration strategy, and (iii) citations to the validation results, thereby establishing the reliability of the reference BMD values used for evaluation.
2) Rationale for using only EOS for muscle mass and BMD estimation without combining CT, X-ray, or DXA (Reviewer #3) There is a clinical demand for using EOS exclusively, without relying on CT, X-ray, or DXA (at least at a hospital where our collaborative orthopedic surgeon practices). At the hospital, EOS is used as a periodical routine clinical examination. If EOS covers muscle mass and BMD estimation, clinical efficiency will be improved. Although this study focused on the hip, the same network can be used to estimate individual muscle volumes across the entire body. Whole-body CT is rarely prescribed, and adding a standing long-length radiograph to EOS prolongs workflow and raises dose. DXA is inexpensive but yields only an aggregate measure of lean mass. It will enable monitoring of region-specific sarcopenia that DXA cannot capture. Standing EOS is common in deformity and arthroplasty clinics, giving load-bearing views that DXA lacks. Leveraging these scans, our method provides muscle-bone metrics to guide rehabilitation and therapy without requiring additional imaging.
3) Limited EOS–CT sample size and generalizability (Reviewer #1) We agree that scale matters. To mitigate overfitting we (i) pre-trained on 600 CT–X-ray image pairs, (ii) used 5-fold cross-validation, (iii) analysed reproducibility with the pre/post-op pairs, and (iv) performed an ablation on fine-tuning set size demonstrating stable performance from only 40 EOS–CT pairs (Fig. 4). These experiments suggest the method’s robustness under realistic data scarcity. We have launched a multi-center collaboration and will report external-site validation in a future extension; this plan will be discussed in the “Limitations and Future Work” section.
4) Perceived limited novelty due to reuse of earlier modules (Reviewers #2 & #3) While we inherit segmentation, registration, and estimation backbones from [6, 9] and our prior works [4, 5], adapting them to EOS required non-trivial contributions, including fine-tuning a pre-trained model and evaluating accuracy on both BMD and muscle mass estimation from a single projection. The ablation study (Table 2) shows that these adaptations yield statistically significant gains over existing methods. We believe that this integrated, EOS-specific pipeline constitutes a significant methodological and clinical advancement.
5) Potential benefit of adding the lateral EOS view (Reviewer #3) We concur that incorporating the lateral projection is a promising approach. We restricted this first study to the AP view to test feasibility in settings where legacy datasets include only a single projection. The Discussion in the final submission will outline an extension with dual-branch encoders and cross-view attention fusion as an explicit future direction, acknowledging the reviewer for their insight.
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- If your recommendation is “Provisional Reject”, then summarize the factors that went into this decision. In case you deviate from the reviewers’ recommendations, explain in detail the reasons why. You do not need to provide a justification for a recommendation of “Provisional Accept” or “Invite for Rebuttal”.
N/A
- 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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A
Meta-review #2
- 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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
The paper get mixed reviews (2 of the reviewers have low confidence). After reading the paper, reviews, and rebuttal, I found the paper have limited technical contribution, lack of comparison to relevant methods. It is a full application work.
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. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A