List of Papers Browse by Subject Areas Author List
Abstract
Aging is complex and heterogeneous. A person’s chronological age (CA) cannot reflect individual aging trajectories which prompted investigations for biological age (BA). Accurate estimation of BA therefore requires to account for the combined influence of structural, functional, and physiological factors on the aging process. BA progresses heterogeneously across organs, motivating models that integrate multiple data types to capture this complexity. Existing BA models however often rely on single organs or limited biomarkers, restricting holistic assessment. Overcoming these constraints requires multimodal models that better reflect the distributed nature of aging. We propose a multimodal contrastive regression framework that jointly learns from MRI and structured clinical variables to estimate organ-specific BA. A contrastive regression loss structures the latent space to reflect continuous age differences both within and across modalities. Applied to a large population cohort, this approach produces well-structured latent embeddings, improves age estimation relative to unimodal systems, and reveals characteristic age-gap patterns in healthy and diseased subgroups. The results demonstrate that combining MRI with complementary tabular features strengthens BA estimation and supports a comprehensive multi-organ view of aging.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/3491_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/lab-midas/multimodal-age
Link to the Dataset(s)
N/A
BibTex
@InProceedings{EckVer_Multimodal_MICCAI2026,
author = { Ecker, Veronika AND Förster, Tim AND Gatidis, Sergios AND Küstner, Thomas AND Yang, Bin},
title = { { Multimodal Contrastive Regression for Organ-Resolved Biological Age Prediction } },
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 introduces a method for organ-wise prediction of biological age. The data used is multimodal - both tabular and imaging data from UK biobank. The method uses a contrastive regression loss
- 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 very clear and easy to follow. The experiments are well structured and very interesting. They include feature domain visualizations, comparison to other methods, experiments on “organ age” for healthy vs subjects with disease, time-to-event and correlations between estimated age of different organs. A discussion on limitations is included.
- 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 setup, including methodology, is rather straight-forward. The lack of ground truth (for biological age) poses challenges, which limits the experiments. However, this is handled as well as possible.
- 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 has provided an anonymized link to the source code, dataset, or any other dependencies.
- 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 is well-written and interesting, but the chosen problem and methodology do not pose major scientific challenges, in my opinion.
- Reviewer confidence
Confident but not absolutely certain (3)
- [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.
My concerns about lack of novelty and ground truth remain (these were also noted by the authors in the rebuttal), and my assessment “weak accept” remains.
Review #2
- Please describe the contribution of the paper
The paper introduces a multimodal contrastive regression framework for estimating organ-specific biological age (BA) by jointly leveraging MRI and structured clinical data. Its main contribution is extending contrastive learning to regression by designing a loss that organizes the latent space based on continuous age differences, enabling meaningful ordering of samples across and within modalities. By integrating complementary data sources, the approach improves age prediction across multiple organs and facilitates analysis of organ-specific aging patterns and disease-related age gaps.
- 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 an approach that combines a meaningful clinical task with a clean methodological contribution and strong empirical validation on large-scale data. The strengths are as follows:
(1) The paper targets organ-specific biological age, a clinically relevant extension beyond global aging biomarkers.
(2) Effectively combines MRI and tabular data, showing consistent gains over unimodal baselines.
(3) Uses large-scale UK Biobank data and demonstrates clinically meaningful age-gap associations with disease
- 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 paper is well-motivated, but a few limitations reduce the strength of its contributions, particularly in novelty and evaluation.
(1) Although the paper is motivated by BA differing from CA, the model is trained under the assumption that BA ≈ CA in a healthy cohort, which is a common but imperfect proxy. This may introduce bias due to undiagnosed conditions and aging variability, potentially limiting biological interpretability.
(2) The contrastive regression formulation builds on prior work extending contrastive learning to continuous targets, and multimodal contrastive learning with imaging and tabular data has also been explored, making the contribution largely incremental.
(3) The evaluation lacks comparison with stronger or more recent models, such as transformer-based multimodal architectures (e.g., UNITER) or vision–language / multimodal foundation models (e.g., CLIP, BLIP), as well as medical imaging backbones, relying mainly on ResNet baselines and unimodal ablations.
(4) Validation is primarily based on predicted age gaps (PAG) and their statistical association with disease presence or future risk. While such correlations are suggestive, they remain indirect proxies and do not provide a direct measure of biological age accuracy, since no ground truth BA exists. In particular, improvements in PAG separation may reflect correlation with known risk factors or confounders already present in the input features, rather than true aging signals. Furthermore, the evaluation does not assess whether the predictions lead to actionable clinical decisions or improved outcomes, limiting the evidence for real clinical utility.
- 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 has provided an anonymized link to the source code, dataset, or any other dependencies.
- 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 addresses an important and clinically relevant problem with a solid multimodal formulation and strong large-scale validation. The contrastive regression approach is well-motivated and shows consistent improvements, along with meaningful disease associations. However, the methodological novelty is somewhat incremental, the evaluation lacks stronger modern baselines, and biological validation remains indirect. Overall, the strengths outweigh the weaknesses, placing it slightly above the acceptance threshold, pending clarification in rebuttal.
- Reviewer confidence
Confident but not absolutely certain (3)
- [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 have clarified most of the concerns, and I am happy to retrain my previous assessment of the paper on the grounds of the proposed important and clinically relevant problem with a solid multimodal formulation and strong large-scale validation.
Review #3
- Please describe the contribution of the paper
The paper presents a multimodal framework for biological age prediction that integrates MRI images from multiple organs with associated tabular clinical data. Results suggest that this multimodal design improves biological age estimation and helps uncover clinically meaningful patterns of organs aging.
- 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.Novel multimodal framework combining organs MRI with relevant structured variables for BA modeling.
2.Instead of using the standard InfoNCE loss, which is primarily designed for classification contrastive learning, the paper proposes a modified version that is better suited for the regression setting.
3.Large UK Biobank cohorts provide a reasonably strong basis to evaluate the proposed framework.
4.The predicted age gap shows clinically relevant information in both existing and future condition groups.
- 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 encoder for tabular data is five-layer MLP. This seems too complex for structured data, a simpler embedding would likely be more appropriate.
2.The assumption that biological age can be approximated by chronological age in the healthy group is too strong and insufficiently justified, since even healthy individuals may still show large variation in biological age.
3.The experimental protocol is unclear. The health cohort is split into 70/30 for train and test sets. There is no validation. Also the paper mentions grid search for kernel parameters, so the model results might rely on test data rather than on a properly separated validation process.
- 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 has provided an anonymized link to the source code, dataset, or any other dependencies.
- 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 paper is interesting, but there are two problems: 1.An insufficiently justified assumption that CA approximates BA in healthy subjects. The CA is treated as label during training. 2.Unclear experimental protocol that does not clearly separate training, validation, and testing.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the authors’ rebuttal, please state your final opinion of the paper.
N/A
- [Post rebuttal] Please justify your final decision from above.
N/A
Author Feedback
We thank the reviewers (R) for their thorough and constructive feedback. The two primary concerns (C) raised are: (1) novelty, with reviewers questioning whether the contribution is sufficiently beyond prior work in contrastive regression and (2) lack of ground truth, with reviewers noting that using chronological age as a proxy for biological age is a strong and imperfectly justified assumption. We address these and the remaining concerns below. (1) NOVELTY (R1-C1, R2-C2) The novelty operates on two levels. First, we present the first multiorgan, multimodal framework for organ-specific biological age estimation, extending prior single-organ and imaging-only approaches to a whole-body setting across six organs with heterogeneous imaging and organ-specific tabular data. Second, we extend contrastive regression to a multimodal continuous-target setting by jointly learning age-ordered imaging and tabular representations. The work addresses a previously unexplored problem through a targeted extension of existing methods, supported by consistent empirical improvements. (2) LACK OF GROUND TRUTH (R1-C2, R2-C1, R3-C2) Reviewers note that assuming BA ≈ CA in a healthy cohort is imperfect and that even healthy individuals may show large variation in biological age. While chronological age in a healthy subcohort is widely used as a proxy for normative biological age, we agree this is a key limitation, as discussed in the Limitations section. For each organ, we excluded participants with known organ-specific diseases and conditions with systemic effects (e.g. type II diabetes). However, absence of recorded disease does not guarantee truly healthy biological aging, as undiagnosed or unrecorded conditions may still be present. While we thoroughly examined all available clinical and diagnostic information in the database, it is conceivable that relevant conditions were unreported, undiagnosed, or not measured. Since these confounders are not known, this remains the best available approximation (R1-C2). Future studies will further investigate potential confounders. (3) MISSING BASELINES (R2-C3) R2 notes the evaluation lacks stronger recent models such as transformer-based or vision-language architectures. Our focus was to isolate the effect of the proposed multimodal contrastive formulation rather than introduce larger pretrained models, which may add confounding factors and are not tailored to biological age estimation. Investigating the impact of architecture choices and medical foundation models is part of our ongoing work, where we are developing transformer-based multimodal fusion and will specifically examine how architectural inductive biases affect biological age estimation. (4) CONFOUNDERS (R2-C4) R2 raises the concern that PAG improvements may reflect correlation with known risk factors rather than true aging signals. We agree that PAG-based validation is indirect, since no ground truth biological age exists. Importantly, the training objective optimizes only for age-related information, not for known risk factors or disease status. We therefore argue that the observed PAG separation reflects aging-related signals that are inherently associated with disease risk, although residual confounding cannot be fully excluded. As this is a first step toward multimodal biological age prediction, we agree that additional analyses will be needed to assess actionable clinical utility and causal dependencies in future work. (5) TABULAR ENCODER (R3-C1) R3 argues that a five-layer MLP for tabular data is overly complex. A simpler single-hidden-layer MLP was initially tested but did not sufficiently preserve age-ordering structure in the tabular embeddings. The deeper encoder produced more stable latent representations and improved alignment and prediction performance. (6) EXPERIMENTAL PROTOCOL (R3-C3) R3 notes the protocol is unclear, with no explicit validation set and potential leakage into hyperparameter selection. We apologize for the missing
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.
While the reviewers agree that the proposed methodology is sound, they have concerns about its novelty, experimental setting, and its evaluation against more recent models. Thus, a rebuttal from the authors is required.
- 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.
Despite some concerns regarding novelty and experimental settings, the reviewers agree that the method is sound, the experiments are thorough, and the problem is important to the field. Then, I suggest acceptance.
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.
This paper utilizes multimodal data for organ-resolved biological age prediction and conducts large-scale validation. Although the experimental setup remains somewhat unclear (as acknowledged by the authors), the rebuttal successfully addressed the majority of the reviewers’ concerns. Therefore, concurring with the assessments of R1 and R2, I recommend acceptance.
Meta-review #3
- 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.
Authors did not fully address the Reviewers’ comments in the rebuttal. For example, the rebuttal reads: “R3 notes the protocol is unclear, with no explicit validation set and potential leakage into hyperparameter selection. We apologize for the missing”, but does not address the criticism per se. Other statements in the rebuttal also reveal the shortcomings of the work presented, leaving the recommendations proposed by the reviewers for future work.