List of Papers Browse by Subject Areas Author List
Abstract
Reconstructing 12-lead electrocardiogram (ECG) signals from single-lead wearable recordings bridges the gap between long-term monitoring for critical paroxysmal events and the clinical gold standard required for accurate diagnosis. However, existing methods treat this ill-posed inverse problem as a generic regression task, failing to resolve physiological ambiguities and consequently reconstructing toward population averages. To address these problems, we propose MetaFormer, an efficient reconstruction network that utilizes demographic characteristics to enforce patient-aware physical consistency. We introduce a meta tokenizer to parse heterogeneous clinical metadata into a unified semantic manifold, which establishes patient-specific physiological context to guide the reconstruction. To resolve the cross-lead spatial ambiguity, a metadata-guided Transformer aligns latent representations of the reconstructed signal with individual physiological baselines. Experimental results on two large-scale benchmarks demonstrate that MetaFormer outperforms other state-of-the-art methods while reducing computational cost by ~73%, and closely approaches the performance of original 12-lead ECG signals on downstream arrhythmia classification. The code is available at: https://github.com/xych3/MetaFormer.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/1866_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/xych3/MetaFormer
Link to the Dataset(s)
N/A
BibTex
@InProceedings{XieYan_MetaFormer_MICCAI2026,
author = { Xie, Yanchong AND Chen, Jiepeng AND Li, Zefeng AND Zheng, Kai AND Huang, Haoyang AND Zheng, Minglong AND Cui, Mingyue AND Huang, Kai},
title = { { MetaFormer: Efficient Metadata-Guided Transformer for Patient-Aware 12-Lead Electrocardiogram Reconstruction } },
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
This paper proposes a novel method called MetaFormer, which incorporates patient metadata as physiological priors to guide a Transformer-based model for reconstructing 12-lead ECG signals from a single-lead input. By addressing the physiological ambiguity inherent in this ill-posed inverse problem, the method improves reconstruction accuracy and avoids the tendency of existing approaches to regress toward population averages. Experimental results demonstrate that MetaFormer outperforms state-of-the-art methods in both reconstruction quality and computational efficiency, while achieving performance on downstream arrhythmia classification tasks that is close to that of real 12-lead ECG signals.
- 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.The problem is defined and of practical significance. Bridging the gap between single-lead wearable devices and clinical 12-lead ECG systems is an important and realistic challenge with strong application value. 2.The paper introduces patient metadata as physiological priors, rather than treating the task as a simple regression problem, which is a meaningful and well-motivated design choice. 3.The paper is well-structured, and the figures are clear and intuitive, making the methodology easy to understand.
- 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 experimental comparisons are not sufficiently comprehensive, as only a limited number of baseline methods are included. 2.The paper does not sufficiently justify the selection of the four metadata attributes, nor does it analyze their individual contributions or explore whether other clinically relevant factors could further improve performance
- 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 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 a problem of clear practical importance and proposes a reasonably novel and well-motivated approach.Reconstructing 12-lead ECG signals from single-lead wearable data is a meaningful and clinically relevant task. The idea of incorporating patient metadata as physiological priors to alleviate ambiguity in reconstruction is intuitive and represents a valuable improvement over standard regression-based methods. However, there are several limitations that prevent a stronger recommendation. First, the experimental evaluation is not sufficiently comprehensive, as only a limited number of baseline methods are considered, making it difficult to fully validate the claimed improvements. Second, the paper does not provide adequate justification or analysis of the selected metadata attributes, including their individual contributions and whether other clinically relevant features could further enhance performance.
- 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
The author proposes a multimodal network for 12‑lead ECG reconstruction from a single‑lead signal and patient demographic characteristics. The model comprises a signal encoder that derives a latent representation from the single‑lead (lead I) input, a metadata tokenizer that generates tokens from demographic attributes (height, weight, age, sex), and a Metadata‑Guided Transformer that reconstructs the 12‑lead ECG using the metadata tokens as keys and values and the single‑lead latent representation as the query. Learnable null tokens are included in the tokenizer to accommodate missing demographic variables. The model is trained on the PTB‑XL (21,799 recordings) and Chapman–Shaoxing (45,152 recordings) datasets.
The results indicate that the proposed model outperforms state‑of‑the‑art baselines in terms of both reconstruction fidelity and downstream arrhythmia classification performance.
- 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.
- Proposed model show great performance: the proposed method outperforms existing approaches while reducing computational cost and adding only minimal additional burden in the form of patient demographic information.
- The paper is well written, with a clear description of the method and implementation details.
- The downstream (cardiac diseases classification) evaluation is quite insight full and show the model’s strong potential for clinical application.
- 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 author does not report confidence intervals or standard deviations, which are particularly important when the performance differences relative to the baseline are small.
- Although the reported metrics (RMSE and PSNR) are useful for model comparison, it would be more clinically interpretable to present errors in ECG features commonly used in diagnosis, such as P-, Q-, R-, S- and T-wave amplitudes, deflections, and intervals.
- 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.
(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 author proposes a novel approach and sufficiently shows that the proposed model outperform current state of the arts.
- 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 proposes MetaFormer, a metadata-guided Transformer for reconstructing full 12-lead ECG from a single reference lead (Lead I), aiming to mitigate the inherent physiological ambiguity of the inverse problem. The core ideas are a meta tokenizer that embeds heterogeneous demographic attributes (age, sex, height, weight) into robust meta tokens (with sinusoidal encodings for continuous variables, type embeddings, and learnable null tokens for missingness), and a metadata-guided Transformer that injects these tokens via cross-attention to align latent ECG representations with patient-specific priors. On PTB‑XL and CSN, MetaFormer improves reconstruction quality and efficiency over recent methods (EKGAN, mEcgNet, ECGrecover) and yields reconstructions that approach the original 12‑lead signals on downstream arrhythmia classification.
- 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.
This work has several strengths in methodology, validation, presentation, and potential relevance. It introduces a metadata tokenizer that can model heterogeneous patient attributes, preserve ordinal structure in continuous variables, and handle missing information through learnable null tokens and simulated dropout. It also designs a metadata-guided cross-attention mechanism that uses patient-specific priors to alleviate cross-lead ambiguity in ECG reconstruction. The overall model is relatively compact and efficient, combining MorphResBlocks with a lightweight Transformer bottleneck to maintain favorable parameter and FLOP cost compared with baseline methods. The experiments cover two large-scale benchmarks, PTB-XL and CSN, and show consistent improvements across multiple reconstruction metrics, including RMSE, PSNR, and CosSim. The ablation studies further examine both metadata injection strategies and tokenizer variants, while the downstream arrhythmia classification results show relatively small gaps compared with original ECGs, supporting the usefulness of the reconstructed signals. The paper is also generally clearly presented: the motivation is easy to follow, the connection between the problem setting and the model design is reasonable, and the method is described with sufficient detail. Overall, the results suggest that metadata conditioning can be a useful way to improve single-lead-to-12-lead reconstruction while keeping the model efficient, and the framework may also be informative for other metadata-informed biosignal or medical inverse problems.
- 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.
What are the exact splitting protocols for PTB‑XL and CSN (patient-level vs record-level), and are there any repeated patients across train/val/test? Can you report results with strict patient-level splits to rule out leakage? Which metadata fields are actually available per dataset (e.g., do PTB‑XL/CSN include height/weight), and what is the missingness pattern? What p_drop was used, and how sensitive are results to higher missingness and to incorrect metadata? How is the “resample to a fixed length of 512” implemented relative to the original sampling frequencies and 10 s durations? Are you cropping, downsampling, or segmenting beats? Please clarify the temporal resolution seen by the model and its impact on morphology fidelity. Can you provide per-lead reconstruction metrics and error distributions (e.g., limb vs precordial leads), and per-diagnosis performance deltas in the downstream classifier to identify where metadata helps most? How was the classifier based on [8] adapted to 12‑lead inputs, and what were the training details (architecture, optimizer, epochs, splits)? Are the reported AUC/F1 averaged over multiple runs with confidence intervals? Have you tested cross-dataset generalization (e.g., train on PTB‑XL, test on CSN) and robustness to noise or motion artifacts on the input lead? If so, how does metadata guidance affect robustness? Did you consider adding simple physiological constraints (e.g., Einthoven’s law, Goldberger relations) or vectorcardiographic priors as auxiliary losses? If attempted, how did these interact with the metadata-guided attention? What is the contribution of each metadata attribute individually (age, sex, height, weight)? An ablation per attribute would clarify which priors are most informative for reconstruction.
- 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?
This is a relevant and generally well-motivated paper that addresses an important limitation of single-lead wearables by incorporating patient-aware priors into 12-lead reconstruction. The metadata tokenizer and cross-attention design are reasonable and efficient, and the method shows consistent improvements over baseline approaches on two benchmarks, with downstream diagnostic results that also support its practical value. To further strengthen the work, it would be helpful to improve methodological rigor, including clearer patient-level split protocols, statistical significance testing, and per-lead analysis. The paper could also benefit from more robustness and cross-dataset generalization experiments, as well as additional discussion of physiological constraints and time–frequency objectives. Clarifying several experimental details, such as resampling and classifier setup, and adding subgroup or fairness analyses would further improve the overall reliability of the study. Overall, MetaFormer provides a useful step toward patient-aware ECG reconstruction and shows reasonable practical relevance, though several aspects could be strengthened to make the contribution more convincing.
- 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
N/A
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.
It is nice paper! Would be interesting to explore the contribution of each meta info, such as gender, age, hight, weight, etc.