List of Papers Browse by Subject Areas Author List
Abstract
Spine aging is a complicated process shaped by pathologies, genetic factors, and lifestyle influences. Radiologists routinely use MR images to assess the spinal health of patients in different age brackets. Quantifying spinal health as an organ age would allow ranking and monitoring of patients within the same and across different demographics. However, spine age estimation has been limited to classical machine learning methods which suffer from high error rates and a lack of interpretability. Moreover, inherently explainable state-of-the-art models in organ age estimation, such as prototypical networks, are limited to 2D and are not extendable to repeated prototype labels. This is important as organs typically degenerate in different ways as a result of aging. We propose ProgreSpine, the first deep-learning-based 3D spine age estimation model based on prototypical regression with a loss specifically tailored to repeated prototype labels. We trained and tuned our proposed model on a large dataset of 9542 samples and performed a thorough evaluation on 1069 samples to demonstrate improved performance against the state-of-the-art with a mean absolute error of 3.61 years. Furthermore, the results suggest that the model learns the prototypes based on clinical conditions that will facilitate monitoring disease progression with a transparent model. The source code is available at https://github.com/prenuvo/progrespine.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/2085_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/prenuvo/progrespine
Link to the Dataset(s)
N/A
BibTex
@InProceedings{BazRoo_ProgreSpine_MICCAI2025,
author = { Bazargani, Roozbeh AND Basar, Saqib AND Hashemi, Sam AND Khallaghi, Siavash},
title = { { ProgreSpine: Inherently Explainable Prototypical Regression for Spine Age Estimation } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15973},
month = {September},
page = {496 -- 506}
}
Reviews
Review #1
- Please describe the contribution of the paper
The proposed ProgreSpine is a deep-learning-based framework for spine age estimation from whole-spine MRI. Specifically, it advances explainable AI in medical imaging by extending prototypical networks to 3D and designing a novel loss function that supports repeated prototypes for the same age label, allowing the model to capture diverse degeneration patterns. By incorporating optimal transport for fine-grained patch-level matching, ProgreSpine achieves state-of-the-art performance while providing inherently interpretable predictions grounded in clinically meaningful prototypes.
- 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.
Spine age estimation is a novel application with potential utility for disease monitoring and patient stratification.
- 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 primary limitation lies in the technical methodology: the approach is relatively straightforward, and its methodological innovation is somewhat incremental.
- 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
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?
(1) The approach is relatively straightforward, and its methodological innovation is somewhat incremental. From a technical perspective, the proposed method lacks innovation. (2) There are relatively few related works and comparative methods. Even if there are not many related methods in the field of medical segmentation, in fact, many methods in natural scenes can be transferred and applied. Can these methods be compared to increase the richness of the experiment? (3）Is the dataset used in the paper a public or private dataset? If it is a private dataset, does the author consider making it available to the public? (4) Please carefully check all formulas, as there may be inconsistencies.
- 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 #2
- Please describe the contribution of the paper
This paper proposes ProgreSpine, a 3D prototypical regression framework designed for predicting chronological spine age from T2-weighted full-spine MRI. Unlike conventional prototype networks limited to classification or 2D data, this model supports repeated prototypes and optimal transport-based distance matching tailored for regression tasks with shared labels. A customized loss function incorporates age discrepancy weighting and diversity regularization, helping avoid prototype collapse and capture population-level degeneration patterns.
The authors demonstrate the method’s utility on a large-scale dataset of 10,611 MRIs, where ProgreSpine achieves better results than multiple baselines including SFCN and 3D-ExPeRT. The framework is also inherently interpretable, with prediction decisions explicitly attributed to learned prototype instances.
- 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). Targeted adaptation of prototypical regression to 3D spine imaging: The model creatively adapts prototypical learning to volumetric data with label repetition, which is common in biological aging tasks. 2). Loss function design fits task semantics: The loss includes softmin aggregation, OT-based matching, and age-aware regularization to preserve prototype structure and diversity. 3). Inherent interpretability without post hoc mechanisms: Model decisions are traceable through weighted distances to stored prototypes, which can be visualized and directly interpreted. 4). Evaluation on a large and diverse dataset: The authors train and test on over 10k spine MRIs and conduct ablation experiments confirming the contributions of each model component.
- 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). Lack of code release or reproducibility resources: The paper does not include any link to code, data, or pretrained models. This significantly impacts reproducibility and community benefit, especially given the complexity of 3D prototype modeling and optimal transport. 2). Prototype interpretability is under-explored: While Fig. 2 shows a single example, there is no systematic analysis of prototype diversity, clustering patterns, or anatomical meaning, weakening the interpretability claim. 3). Architectural novelty is modest: The overall structure builds on known concepts (e.g., prototypical networks, OT), and the contribution lies more in adaptation and loss design than novel modeling. 4). Single-dataset evaluation limits generalizability: The model is only tested on one internal dataset. There is no external validation across institutions, vendors, or demographics—crucial for clinical translation. 5). Bias toward younger subjects remains unaddressed: The authors note systematic underestimation for subjects aged >80 but do not attempt solutions (e.g., age-aware sampling, weighting strategies).
- 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 provide sufficient information for 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
None.
- 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?
This paper proposes a strongly engineered and clinically relevant adaptation of prototypical learning to spine age estimation. The methodology is solid, the evaluation is large-scale, and the integration of interpretability via prototypes is conceptually sound.
However, the absence of code, combined with limited novelty and partial interpretability analysis, reduces its strength as a community resource. With code and deeper interpretability experiments, this paper would be a clear accept. In its current form, I recommend Weak Accept, contingent on improving transparency and reproducibility post-review.
- 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.
Most of questions have been replied by authors.
Review #3
- Please describe the contribution of the paper
A method to regress the patient age and explain spine conditions from 3D MRI data is proposed. The deep learning method is able to predict the age as a weighted combination of closest prototypes from 3D scan spine ROI and is trained from a large MRI dataset. Assessment of the method is done using 1069 samples and provide 3.61 year MAE difference an R2 correlation of 0.85 between actual and predicted age.
- 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 standard error of estimate shows good performances (correlation R2=0.85) -The explainability is a must for interpretation by radiologists to better figure out the disease progression
- 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.
-As for spine specialist, it could be evident, the interest of predicting the patient age for clinical application purpose should be better stated -The principle of Repeated Prototype has to be better explained (rational, etc…)
- 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
-A visual plot of Bland & Altman is usually done to analyze the regression results to better figure out the residual distribution -Eq.8 is quite standard, if authors are lacking paper space, they can remove it -
Batch size of 2 was used and we accumulated the gradients for 3 iterations before backpropagation.This is not standard training, is the backprop taking the average of the gradients then ? - 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?
This paper brings interesting approach for prototype-based prediction for complex anatomical structure (spine)
- 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’ answers and the revised paper should bring improved scientific contribution to the community, especially in the field of spine age estimation from 3D image supported by explainability of complex spine deformities.
Author Feedback
We thank reviewers R1, R2, and R3 for their constructive comments and for highlighting strengths of the work, including the novel application [R1], solid methodology, large-scale evaluation [R2], good performance, and importance of explainability [R3]. We have addressed your comments in Questions 7, 10, and 12 by numbering them.
Reproducibility [R1.12.3, R2.7.1] Due to institutional restrictions, we are unable to release the dataset. However, we plan to provide the code upon acceptance.
Prototype interpretability [R2.7.2] We have the corresponding radiology reports for our dataset. After embedding report conditions into a vector and UMAP visualization, we found that prototypes cover all clusters in different age brackets, representing different degenerative patterns. We added the figure and, with the published code, addressed R2’s key concerns to strengthen our case for acceptance.
Repeated prototypes [R3.7.2] We observed that each age group has different patterns of degeneration based on the radiology report and how the diseases progress over time, depending on spine region and type of degeneration, such as the number and severity of disc bulges or osteophytes in the lumbar. Thus, we decided to have multiple prototypes to represent each age. The effectiveness of this idea is shown in Table 1, as it reduced the MAE by 1 year. We added a discussion to the revised paper.
Novelty [R1.12.1, R2.7.3] While our method builds on prior work, this is the first 3D extension of prototypical regression. We also introduce the first deep learning framework for spine age estimation, evaluated on a large and diverse clinical dataset. Our repeated prototype loss further enables modeling of intra-age variability. This offers an effective and interpretable solution to the unique challenges of spine age prediction, which we believe is of interest to the community.
Comparative methods [R1.12.2] We compared our approach against strong baselines, including prototype-based models and CNN regressors using MSE / ordinal losses. To our knowledge, there are no existing prototypical regression methods readily applicable to 3D spine age estimation task. While other regressors for age prediction exist, they rely on post-hoc interpretability (e.g., attention maps), whereas our method offers built-in interpretability through learned prototypes.
Clinical application [R3.7.1] We revise the paper to better motivate biological spine age: it enables more accurate risk stratification by identifying patients whose degeneration is advanced relative to their chronological age. It can also inform clinical decision-making by assessing disease progression and provide patients with an individualized measure to guide long-term care.
Generalization [R2.7.4] While our evaluation is limited to a single organization, the dataset includes over 10000 MR images from 10 clinics using both Philips and Siemens scanners, covering diverse regions and demographics. This internal variability captures a range of real-world conditions and improves robustness. We agree that external validation across institutions is important for clinical translation, and we will pursue it in future.
80+ underestimation [R2.7.5] We explored reweighting strategies prior to submission but observed limited gains due to the extreme class imbalance, only 25 subjects aged over 80 out of 10611 (0.24%). Notably, all baseline methods also underperform in this range, highlighting a broader challenge. We clarify the discussion in future work and believe it can be best addressed through targeted data collection.
Formulas [R1.12.4, R3.10.2] We updated the formulas and fixed inconsistencies.
Bland & Altman plot [R3.10.1] We have generated the plot and added it to the paper.
Accumulation of gradients [R3.10.3] Yes, the gradient accumulation strategy is functionally equivalent to using a batch size of 6. This technique is commonly adopted when GPU memory constraints prevent the use of large batch sizes directly.
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.
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 #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