List of Papers Browse by Subject Areas Author List
Abstract
Accurate lesion tracking in temporal mammograms is essential for monitoring breast cancer progression and facilitating early diagnosis. However, automated lesion correspondence across exams remains a challenges in computer-aided diagnosis (CAD) systems, limiting their effectiveness. We propose MammoTracker, a mask-guided lesion tracking framework that automates lesion localization across consecutively exams. Our approach follows a coarse-to-fine strategy incorporating three key modules: global search, local search, and score refinement. To support large-scale training and evaluation, we introduce a new dataset with curated prior-exam annotations for 730 mass and calcification cases from the public EMBED mammogram dataset, yielding over 20000 lesion pairs, making it the largest known resource for temporal lesion tracking in mammograms. Experimental results demonstrate that MammoTracker achieves 0.455 average overlap and 0.509 accuracy, surpassing baseline models by 8%, highlighting its potential to enhance CAD-based lesion progression analysis. Our dataset will be available at https://gitlab.oit.duke.edu/railabs/LoGroup/mammotracker.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/4723_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://gitlab.oit.duke.edu/railabs/LoGroup/mammotracker
Link to the Dataset(s)
Our dataset is built on the EMBED dataset: https://gitlab.oit.duke.edu/railabs/LoGroup/mammotracker
EMBED dataset: https://registry.opendata.aws/emory-breast-imaging-dataset-embed/
BibTex
@InProceedings{LiuXua_MammoTracker_MICCAI2025,
author = { Liu, Xuan and Ren, Yinhao and Ryser, Marc D. and Grimm, Lars J. and Lo, Joseph Y.},
title = { { MammoTracker: Mask-Guided Lesion Tracking in Temporal Mammograms } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15963},
month = {September},
}
Reviews
Review #1
- Please describe the contribution of the paper
The authors propose MammoTracker, a novel framework for accurate lesion tracking in temporal mammograms. They also release the largest annotated temporal mammogram dataset to date.
- 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 study introduces a novel dataset that could be useful to the community.
- 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 appears disorganized and challenging to follow; for example, the first figure presented focuses on results rather than providing context or presenting the methodology.
Section 3 (“Method”) would benefit from clearer structure and a more explicit separation with implementation details. Several important aspects are missing, including: (1) a better explanation of the global search strategy, ideally with references to existing work in medical image registration; (2) clarification on the use of the pre-trained MobileNetV2, specifically, whether 3-channel inputs are used and, if so, a justification for this choice; and (3) a description of the training procedure for the entire framework, whether it is end-to-end or involves freezing components such as the local search. It would also be helpful to annotate the losses computed at different stages in Figure 3 to enhance clarity. Reference [18] corresponds to SimpleITK, which is a general-purpose image processing toolkit rather than a state-of-the-art lesion tracking method. Including it as a baseline for performance comparison is somewhat misleading and weakens the experimental evaluation, as it does not represent a competitive or specialized approach in the context of lesion tracking.
- Please rate the clarity and organization of this paper
Poor
- 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 authors claimed to release the source code and/or dataset upon acceptance of the submission.
- 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.
(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?
The paper is unclear and its quality unsufficient. Please, refer to the major weaknesses to see the detailed reason of the reject.
- 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 paper presents a deep learning approach for lesion tracking in temporal mammograms of the same view. The method is based on a global search (using an affine registration), a local search based on embdding similarity using cross-correlation and a score refinement. The paper is well-written and clearly structured. The methodological contribution is incremental but effectively tailored to the specific challenge of mammography lesion tracking, which adds to its relevance. The use of a publicly available dataset (EMBED) enhances reproducibility. Additionally, the authors enriched the dataset by providing ground truth annotations for lesion tracking across a large number of cases.
- 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.
- Well presented and structured paper.
- Interesting proposal tailored approach including several steps: affine registration, local search using a local correlation loss function and score refinement.
- Good experimental section with comparison with 3 soA approaches using an large publicly available dataset EMBED Publication (not details given) of an extended GT data for lesion tracking which could be interesting for further research on mammography.
- 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 methodological novelty, and a very specific method, difficult to make it generalisable to other tasks and domains. Some steps are adhoc (score refinement) and not exhaustively evaluated.
- Limited experimental aspects: no evaluation is done on the accuracy of the global search, successfull / robustness tracking is not clearly defined, the global search and score refinement step are not evaluated in the ablation study.
- Limited discussion on the clinical applicability of the approach and if the improvements presented are of clinical relevance and robust enough to be applicable to additional datasets given the supervised nature of the approach.
- 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?
The overall recommendation is based on existing weaknesses. The paper has some merit regarding the methodological contribution but some aspects are not fully evaluated and justified. The publication of the enhanced ground truth for lesion tracking is also interesting but no details are given. Results show the validity of the approach compared to the state of the art but the discussion on downstream tasks and clinical application is somewhat limited.
- 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 a novel method for tracking lesions across longitudinal mammograms, focusing on both masses and calcifications. The proposed framework consists of three main components:
- A global search module using affine registration
- A local search module that performs mask-guided, anchor-free search
- A score refinement module that leverages similarity learning to improve matching
- 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.
- Lesion tracking is a critical step for longitudinal breast cancer analysis and early detection, and this work makes a meaningful contribution in that direction.
- The proposed approach is comprehensive, combining global and local alignment with refinement. The ablation study helps highlight the value of some of the implementation details
- The authors also provide annotated lesion-tracking data for the EMBED dataset, which is a valuable contribution to the community.
- The method is compared against two relevant baselines, including affine registration and anchor-free tracking. The proposed pipeline appears to integrate and improve upon these approaches.
- 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 performance gain for calcification tracking is relatively modest compared to masses. This is likely due to the small size and low contrast of calcifications, which makes them difficult to detect and track accurately. It would be beneficial if the authors could elaborate further on failure cases and possibly explore calcification-specific enhancements.
- As the authors mentioned in their future work, it would be valuable to quantify how tracking improves downstream detection performance. Including even a preliminary analysis or discussion of this could strengthen the paper’s overall impact.
- 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 authors claimed to release the source code and/or dataset upon acceptance of the submission.
- 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.
(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 work makes a good contribution, particularly through its release of annotated tracking data and integration of multiple tracking strategies. The approach is promising.
- 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
We thank the reviewers for their valuable comments. In this work, we propose MammoTracker, a lesion tracking framework that integrates affine registration for global search (Section 3.1), mask-guided anchor-free tracking model for local search (Section 3.2), and a score refinement module to further enhance performance (Section 3.3). The local search model is built on a pre-trained MobileNetV2 backbone with 3-channel inputs. Each step of the framework is trained independently. Overall, MammoTracker achieves an average overlap of 0.455 and an accuracy of 0.509, outperforming baseline models by 8%. In this work, SimpleITK is used only for implementing affine registration in the global search step.
The goal of this work is to facilitate lesion tracking across temporal mammograms, which helps increase lesion-level annotations in longitudinal studies. This is a critical step toward analyzing lesion growth and morphological changes over time. By incorporating comparisons with prior images, our approach may also contribute to downstream tasks such as lesion detection, classification, and risk prediction in CAD systems.
As part of our ongoing work, we are expanding our experiments to additional datasets, including OPTIMAM and a Duke private dataset. We will further evaluate MammoTracker’s performance on these datasets and plan to extend our framework to support downstream tasks, including both classification and risk prediction.
Meta-Review
Meta-review #1
- Your recommendation
Provisional Accept
- 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