List of Papers Browse by Subject Areas Author List
Abstract
Body composition assessment (BCA) provides detailed information about the distribution of different tissue types in the body, enabling more precise characterization of individuals than BMI or weight alone.
Consistent and frequent BCA would be valuable for personalized medicine, but the gold standard methods for BCA, such as CT and MRI, are only practical for opportunistic monitoring of patients with clinical indications for imaging and are not suitable for routine use in the general population. Here, we consider an imaging modality which is not currently used in medical applications: millimeter wave (mmWave) radar. Commonly used in security settings, mmWave scans enable fast, non-intrusive, and privacy-preserving reconstruction of full body shape without the need to remove clothing. To demonstrate the feasibility of fast and convenient BCA from mmWave scans, we present a method for BCA value regression using a multi-task learning strategy that leverages synthetic mmWave-like point clouds derived from clinical imaging and parametric human models. We evaluate the model on a pilot cohort of real mmWave scans with bioimpedance-derived body fat measurements, supporting the feasibility of estimating VAT and body fat percentage (BFP) from mmWave data acquired through clothing in a standing posture. We find that the model can predict VAT and BFP with a mean absolute error of 1.0 L and 3.2%, respectively, demonstrating the potential of mmWave scanning for routine BCA in a wide range of settings.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/5425_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)
NMDID dataset: https://nmdid.unm.edu
HIT dataset: https://hit.is.tue.mpg.de
BibTex
@InProceedings{SenMir_Nonintrusive_MICCAI2026,
author = { Senne, Miriam AND Killeen, Benjamin D. AND Wang, Tony Danjun AND Navab, Nassir},
title = { { Non-intrusive Body Composition Assessment from Full-Body mmWave Scans } },
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 main contribution of this paper is a feasibility study of a novel imaging modality for body composition assessment. The authors develop an AI-based solution, leveraging vision transformers, to estimate visceral adipose tissue (VAT) and body fat percentage (BFP) from mmWave scanner data. This approach has the potential to enable opportunistic screening for metabolic diseases during routine body assessments in the general population.
The proposed method is fast and non-invasive. The authors evaluate their approach using data acquired from an mmWave scanner and compare the estimated VAT and BFP values against measurements obtained from bio-impedance body fat analysis.
- 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 primary strength of this paper is the introduction of a novel imaging-based approach for body composition assessment using mmWave data. This represents an innovative alternative to traditional modalities such as CT and MRI, offering a solution that is potentially lower-cost, faster, and non-invasive, while still aiming to capture clinically relevant metrics such as visceral adipose tissue (VAT) and body fat percentage (BFP). This makes the approach promising for large-scale, opportunistic screening of metabolic diseases.
Another key strength is the strategy used to address the lack of paired mmWave and CT/MRI datasets. The authors leverage prior information from publicly available full-body CT and MRI datasets to guide the learning process.
The paper also tackles the challenge of differing acquisition conditions (e. g. , subject positioning differences between mmWave and CT/MRI).
From a methodological perspective, the use of vision transformer–based models for estimating VAT and BFP from mmWave and prior information derived from MRI/CT datasets is a relevant choice. Applying such architectures to this type of data and task demonstrates originality and extends the application of transformer-based models to a less explored imaging domain.
The work could enable non-invasive and accessible body composition assessment, facilitate early detection and monitoring of metabolic conditions in broader populations.
- 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 do not provide validation against gold-standard measurements such as paired CT or MRI scans. Instead, the evaluation relies on indirect comparisons, which limits confidence in the accuracy of the estimated VAT and BFP.
The test cohort appears to be small, which may raise concerns about the generalisability of the results.
- 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.
(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 abstract presents a novel approach with clear potential for impact at a population scale. The use of a non-invasive and potentially low-cost imaging modality for estimating body composition is promising, as it could enable broader and more accessible screening for metabolic diseases.
- 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 proposes a framework for Body Composition Analysis (BCA) leveraging mmWave information. Built upon a Point Transformer backbone, the proposed network employs a multi-task optimization strategy to derive BCA metrics directly from the synthesized point clouds. Compared to conventional BCA measurement approaches, this paper is non-invasive and privacy-preserving.
- 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 proves the potential of using mmWave scanners for BCA measurement.
- 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.The network design appears to lack sufficient novelty, as the architecture relies on a standard Point Transformer backbone with multiple classification heads. 2.The paper lacks a comparison with existing methodologies. It is highly recommended to identify and include other relevant BCA methods to provide a clearer context for evaluating the proposed approach. 3.Table 1 does not present the main results of the proposed method. 4.The paper suffers from a lack of comprehensive experimental analysis. a) The reason why mmWave point clouds can estimate VAT and BFP is not well justified, which raises concerns about the generalization capability of the proposed method—i.e., whether the method performs well on unseen subjects (not included in training). b) The quality of the collected mmWave points is essential for subsequent measurements, but the authors did not analyze its impact. c) The material of clothing should affect the mmWave points, yet no relevant experiments were conducted to evaluate this effect. d) It is important to report the distribution of body shape, age, gender, and other relevant information, as the diversity of human subjects is crucial for demonstrating the effectiveness of the proposed method. e) Other conditions—such as the motion of the human subject, the distance between the front and back mmWave scanners, and long-term measurements of the same subject—are also important impact factors, but they are not mentioned.
- 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.
(2) Reject — should be rejected, independent of rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
While the paper demonstrates the potential of using mmWave scanners for BCA measurement, the lack of necessary validation experiments undermines the credibility of the method’s effectiveness and practicality.
- 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
Review #3
- Please describe the contribution of the paper
This paper introduces a methodology to determine if mmWave radar data can be successfully used for Body Composition Assessment (BCA). The central challenge the authors face is a significant data gap: there is almost no available data where subjects have undergone both a routine mmWave scan and a medical scan like an MRI or CT for body composition. To overcome this lack of paired data, the authors developed a pipeline to synthesize mmWave-like point clouds from existing medical imaging. However, this process introduces a new problem: medical scans are taken while a patient is lying down (supine), whereas mmWave scans are performed while standing. This change in posture shifts how body fat and tissue are distributed. To bridge this gap, the authors utilize the Skinned Multi-Person Linear (SMPL) model, which allows them to adapt these medical body shapes into realistic standing poses. The technical core of the paper is a Multi-Task Learning (MTL) framework. By using a single point cloud encoder (Point Transformer V3) shared across multiple regression heads, the model is forced to learn a, shared representation of the human body. These specialized heads simultaneously predict three heads: one for internal fat volumes, one for height, and one for anthropometricl metrics like waist, chest, and hip circumference. The authors validate their approach using a real-world mmWave cohort where the results are compared against bioimpedance (BIA) 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 authors solve the data problem by turning MRI/CT scans into 3D point clouds. Using the SMPL model to fix the pose difference between lying down and standing. The methodology demonstrates that Multi-Task Learning (MTL) provides significant advantages in data-constrained environments. By predicting anatomical measures (such as height and waist circumference) alongside internal fat volumes and showing improved performance in correlation between VAT and BF when compared to single-task baselines. The paper includes Masked Sample Loss and dynamic weighted objective as an optimization approach to balance task-specific gradients and ensuring model convergence across targets.
- 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 excluded 41 scans due to extreme body fat values. Given that individuals with high visceral fat are the primary target population for metabolic screening, does this exclusion limit the clinical utility of the system? It would be beneficial to discuss why these anatomical extremes are challenging. In Table 1, the reported SAT and VAT MAE values appear high relative to measurements of typical SAT VAT range[1]. Could the authors provide more context or intuition as to why this is? Lastly, the authors compare a baseline trained only on CT/MRI against the full MTL pipeline. However, the intermediate step of MRI/CT + SMPL or MRI/CT+mmWave is missing.
[1] https://pmc.ncbi.nlm.nih.gov/articles/PMC3636065/
- 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 paper introduces a technically innovative pipeline that effectively uses SMPL models, CT/MRI and Multi-Task Learning to overcome a critical lack of paired clinical data. While the performance on real-world mmWave scans is promising, the exclusion of high-BMI subjects and the absence of intermediate ablation studies make it difficult to verify the model’s clinical robustness. If the authors address these gaps, the work would represent a significant step forward for non-invasive health monitoring.
- 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
Author Feedback
We thank the reviewers for their thoughtful comments. Reviewers highlighted the novelty of the mmWave imaging modality for body composition assessment (BCA) (R1,R2,R3,M2), the contribution of the multi-modal, multi-source synthetic-to-real pipeline (R1,R3,M2), and the strong motivation of providing nonintrusive, accessible imaging for personalized medicine (R1,R2,M2) and opportunistic screening of metabolic diseases (R1).
Novelty (R2): While we agree that the model architecture uses established point cloud encoders, we would argue for the novelty of leveraging mmWave scans for BCA, with the goal of enabling non-intrusive, personalized medicine. This modality has not previously been considered for this task, and as noted by M2, the main contribution lies in the synthetic-to-real pipeline which enables model training.
Comprehensive, longitudinal, gold-standard evaluation (R1,R2.d,e, M2): We share the reviewers’ desire for comprehensive, longitudinal data collection with gold standard ground truth for BCA. Nevertheless, we believe our results, which explore mmWave scans as a medical imaging modality for the first time, prove the potential of this modality, motivating the time and cost investment necessary for large-scale, long-term data collection. We will update our discussion to point toward this future work.
Acquisition variability (R2.b,c,e): We agree that scan variability, e.g. demographics, clothing, and pose, are important for validation. In our real data collection, each participant underwent three consecutive scans to account for pose and noise variability, with predictions averaged across measurements. Participants were clothed in streetwear to emulate casual acquisitions and instructed to stand on fixed floor markers in a standard A-pose. Participants ranged from 21 to 31 years old, from 51 to 108 kg, and consisted of 8 men, 6 women. We will clarify these points in the experimental description.
Baselines (R2.a): While related work estimates BCA based on full-body meshes, this is the first work exploring mmWave scans as a medical imaging modality. As M2 notes, the primary contribution lies in the synthetic-to-real pipeline: CT/MRI-derived point clouds, SMPL-based pose bridging, and masked multi-task learning. We will clarify that our evaluation of these components demonstrates that they make BCA from mmWave scans possible.
Exclusion of outliers (R3): We agree that BCA methods should be performative across a range of body shapes, and the reviewer’s comment points toward a potential limitation of the current pipeline. At the low end (<5% BFP), even “minimally decomposed” samples can carry anatomical artifacts corrupting TotalSegmentator-derived BCA values. At the high end, (>70% BFP), field-of-view truncation made both labels and surfaces impossible to obtain. We will acknowledge these limitations in the CR.
Intermediate ablations (R3): We share the reviewer’s desire to improve our data pipeline by exhaustively isolating the contribution of each component, e.g. by ablating SMPL-based or real data. Nevertheless, we believe our results demonstrate the feasibility of BCA estimation from mmWave scans, an imaging modality never before considered in this context, and the contribution of the multi-headed training pipeline to enable synthetic-to-real training.
Minor comments (R3): We will provide context for SAT and VAT MAE values, which are affected by annotator variability regarding the boundaries of the abdomen.
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.
This submission presents a novel and technically coherent feasibility study on body composition assessment from full-body mmWave scans, addressing an important clinical-monitoring problem through a non-intrusive and privacy-preserving sensing modality. The strengths of the paper are the originality of the application setting and the integrated methodological design used to address the lack of paired mmWave and gold-standard imaging data. In particular, the manuscript combines synthetic mmWave-like point cloud generation from CT/MRI, SMPL-based handling of posture mismatch, and partial-label multi-task learning in a way that is well aligned with the stated problem. The paper also provides initial real-data evidence, including improved pilot-cohort correlations over a simpler single-head baseline, which supports the core feasibility claim. The decisive concerns relate to the strength and scope of the validation. After checking the manuscript, it is clear that the real body-composition evaluation is based on a small pilot cohort and uses bioimpedance rather than paired CT/MRI as reference, which limits the strength of clinical conclusions. In addition, applicability to extreme body-fat cases and robustness across acquisition conditions are not fully characterized. At the same time, some reviewer criticisms are only partially supported. In particular, the concern that the main results are missing from the paper appears to stem from a misunderstanding, since the relevant real-cohort BCA results are presented in Fig. 3, with an explicit explanation for using correlation rather than direct MAE. Likewise, the objection that the work lacks novelty is overstated if judged only at the encoder level, because the main contribution lies in the integrated synthetic-to-real training and domain-bridging pipeline.