List of Papers Browse by Subject Areas Author List
Abstract
Apical periodontitis is a prevalent oral pathology that presents significant public health challenges. Despite advances in automated diagnostic systems across various medical fields, the development of Computer-Aided Diagnosis (CAD) applications for apical periodontitis is still constrained by the lack of a large-scale, high-quality annotated dataset. To address this issue, we release a large-scale panoramic radiograph benchmark called “PerioXrays”, comprising 3,673 images and 5,662 meticulously annotated instances of apical periodontitis. To the best of our knowledge, this is the first benchmark dataset for automated apical periodontitis diagnosis. This paper further proposes a clinical-oriented apical periodontitis detection (PerioDet) paradigm, which jointly incorporates Background-Denoising Attention (BDA) and IoU-Dynamic Calibration (IDC) mechanisms to address the challenges posed by background noise and small targets in automated detection. Extensive experiments on the PerioXrays dataset demonstrate the superiority of PerioDet in advancing automated apical periodontitis detection. Additionally, a well-designed human-computer collaborative experiment underscores the clinical applicability of our method as an auxiliary diagnostic tool for professional dentists.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/1336_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/XiaochengFang/MICCAI2025_PerioDet
Link to the Dataset(s)
https://github.com/XiaochengFang/MICCAI2025_PerioDet
BibTex
@InProceedings{FanXia_PerioDet_MICCAI2025,
author = { Fang, Xiaocheng and Cai, Jieyi and Liu, Huanyu and Zhou, Chengju and Lu, Minhua and Chen, Bingzhi},
title = { { PerioDet: Large-Scale Panoramic Radiograph Benchmark for Clinical-Oriented Apical Periodontitis Detection } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15975},
month = {September},
page = {421 -- 431}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper introduces PerioDet, a detection framework for apical periodontitis in dental panoramic radiographs, along with PerioXrays, while addressing background noise and small target detection through Background-Denoising Attention and IoU-Dynamic Calibration mechanisms.
- 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 creation of the first comprehensive dental benchmark dataset (PerioXrays), novel attention mechanisms addressing background noise, adaptive IoU thresholding for small lesion detection, superior performance over state-of-the-art models, and robust clinical validation demonstrating significant improvements in diagnostic accuracy and efficiency.
- 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 Background-Denoising Attention module appears to have a relatively simple structure. In this context, how can it be interpreted as performing true denoising? Is it reasonable to assume that applying a sigmoid function effectively filters out meaningful features?
-
Regarding the Adaptive IoU Threshold, parameters such as W, H, and especially As could lead to frequent selection of a 0.25 threshold. Wouldn’t this result in rough rather than precise detections? Moreover, this may influence the confidence score p, potentially affecting the training process via the proposed Dynamic Label Assignment. Additional explanation is needed on how this mechanism operates and its overall impact on model learning.
-
- 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?
Please refer the main weaknesses of the paper.
- 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 main contributions of the paper are as follows: The work introductes PerioXrays, the first large-scale, expert-annotated benchmark dataset specifically designed for apical periodontitis detection from panoramic dental radiographs. To address core challenges in this task—namely background noise and small lesion size—the paper also proposes a novel detection framework, PerioDet, which incorporates: Background-Denoising Attention (BDA): A mechanism that enhances feature representation by emphasizing relevant lesion features and suppressing irrelevant background signals. IoU-Dynamic Calibration (IDC): An adaptive thresholding method that adjusts anchor assignment criteria based on lesion size, thus improving detection sensitivity, particularly for small apical lesions. Through experiments and a human-computer collaboration study, the authors demonstrate that PerioDet achieves state-of-the-art performance and holds potential as a clinically useful diagnostic aid for dentists.
- 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 major strengths of the paper are:
Introduction of a Novel, Large-Scale Dataset (PerioXrays): The paper presents PerioXrays, the first large-scale, publicly available dataset focused on apical periodontitis detection from panoramic radiographs. With 3,673 high-resolution images and 5,662 expertly annotated lesion instances, this dataset addresses a critical gap in the field and offers a valuable resource for future research in dental radiography and AI-driven diagnosis.
Clinical-Oriented Design and Annotation Quality: The dataset creation process involved rigorous multi-stage review by four experienced dentists, ensuring high annotation accuracy. The diversity of demographic representation and consistent image resolution (1333×800) make the dataset both clinically relevant and technically robust.
Novel Methodology – Background-Denoising Attention (BDA): The BDA module introduces a clever mechanism to enhance lesion-specific features by modeling channel-wise importance and filtering out irrelevant background, which is particularly challenging in panoramic dental images known for artifacts and low contrast.
Adaptive Detection Strategy – IoU-Dynamic Calibration (IDC): The IDC module innovatively employs adaptive IoU thresholds based on lesion size, improving anchor-label matching, especially for small and subtle lesions. The inclusion of a Dynamic Label Assignment strategy further refines this process during training, enhancing detection accuracy.
Comprehensive Experimental Validation: The method is extensively evaluated on the PerioXrays dataset, with quantitative results supporting its effectiveness. Moreover, the inclusion of a human-computer collaborative study emphasizes the practical clinical feasibility of the proposed approach and its potential as an assistive diagnostic tool for dental professionals.
Clear Problem Framing and Real-World Relevance: The paper is well-grounded in a real clinical problem—accurate detection of apical periodontitis from commonly used imaging modalities—making it a strong example of impactful AI for healthcare.
- 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 major weaknesses are as follows:
Limited Baseline Comparisons and Ablation Scope: While the paper introduces innovative modules (BDA and IDC), the experimental section could benefit from broader baseline comparisons, particularly with recent strong object detection methods that also address small object detection or medical imaging challenges (e.g., YOLOv7, DETR-based architectures, or TransUNet for region detection). The ablation of Large Vision language models on the benchmark can be a good direction and can derive new insighs. Furthermore, while ablation studies are present, a more granular breakdown of how BDA and IDC each affect performance under different lesion sizes or anatomical regions would improve interpretability.
Lack of Generalization Analysis Beyond PerioXrays: The proposed model is evaluated solely on the newly introduced dataset. While the dataset itself is a major contribution, the absence of external validation (e.g., on other public dental datasets like UFBA-UESC or private datasets from different scanners) leaves the generalizability and robustness of the method untested.
IDC Design Justification Could Be Strengthened: The paper introduces a dynamic IoU thresholding scheme, but lacks theoretical or empirical justification for the chosen functional form (e.g., how the dynamic threshold is determined based on box area). Alternative thresholding strategies (such as focal loss–based approaches or auto-learned thresholds) could have been discussed or tested to contextualize the benefit of IDC.
Limited Discussion on Annotation Ambiguity and Inter-Rater Variability: Although annotation quality is high, there is no quantitative analysis of inter-annotator agreement (e.g., kappa scores or overlap metrics), which is crucial given the subtlety of apical periodontitis in X-rays. Such an analysis would help establish the upper-bound of human performance and offer a benchmark for the model.
Underexplored Error Analysis and Failure Modes: The paper presents strong average metrics but lacks a qualitative or quantitative breakdown of failure cases. Insight into typical misdetections (e.g., false positives near dental artifacts or undetected small lesions) would enhance understanding of the model’s limitations and guide future improvements.
Slight Overlap with Prior Attention-Based Work: The proposed BDA module shares conceptual similarities with attention-based denoising modules used in vision transformers (e.g., DeiT, TransFuse) and medical detection models (e.g., Attention U-Net). While this paper adapts the idea to dental imaging, a clearer articulation of how BDA is distinct from or improves upon these existing methods would strengthen its novelty claim.
- 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.
- 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
My additional comments to the Authors:
This paper presents a well-motivated and clinically relevant contribution to the field of dental imaging. The creation of the PerioXrays dataset is timely and valuable, addressing a clear gap in publicly available annotated resources for periodontal disease detection. The proposed architecture—incorporating the Bi-Directional Aggregation (BDA) module and the IoU-aware Dynamic Confidence (IDC) mechanism—is thoughtfully designed, and the empirical results are strong.
That said, there are a few areas where the paper could be further strengthened:
Providing more detailed analysis of failure cases would help contextualize performance and guide future work.
Including external validation or cross-domain testing could reinforce claims of generalizability.
The theoretical underpinnings or motivation behind some of the design choices (especially in IDC) would benefit from further clarification or comparison with alternatives.
In terms of presentation, the paper is well-written and flows logically. Visualizations are clear and helpful, particularly the qualitative comparisons and dataset overview. The work is likely to be of interest to both clinical and technical audiences within the MICCAI community.
I encourage the authors to release the PerioXrays dataset publicly, as it would be a valuable resource for fostering further research in dental AI. Overall, this is a strong submission with promising potential for both academic and clinical impact.
- 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?
I recommend a Weak Accept for this paper. It addresses an important and underexplored clinical task—periodontal disease detection from dental X-rays—and introduces the PerioXrays dataset, filling a clear gap. The proposed architectural components (BDA module and IDC mechanism) are thoughtfully designed and yield consistent performance improvements across multiple metrics. While the methodology is strong, the paper would benefit from deeper theoretical motivation, comparisons with related techniques, and external validation to assess generalizability. Greater transparency around annotation quality would also enhance clinical credibility. Overall, the work is solid and has the potential to inspire future research in this domain.
- 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 #3
- Please describe the contribution of the paper
- Benchmark dataset of panoramic radiographs, specifically curated and annotated for apical periodontistis detection.
- Background-Denoising Attention (BDA) module to handle background noise and an IoU-Dynamic Calibration (IDC) module
- 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.
- Significant dataset contribution
- Clinically relevant problem
- Targeted methological innovation
- Strong experimental validation
- Human-computer collaboration study
- Reproducibility
- 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.
- Dataset diversity details (equipment/protocols) could be expanded
- Human-computer study is relatively small
- Novelty of sub-components (attention, thresholds) within broader literature could be better situated
- 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.
- 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 paper significantly contributes the large benchmark dataset for this task and proposes a novel method PerioDet with strong results addressing key challenges. The clinical relevance and provided code/data support acceptance.
- 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 all three reviewers for their thoughtful evaluation and constructive suggestions. R1-W1: Interpretation of the BDA module performing true denoising. BDA leverages two complementary mechanisms—(a) a learned channel-importance vector zi to suppress background channels, and (b) a pixel-wise similarity map Si to highlight scene-consistent lesion patterns. Together, they act as a data-driven “soft filter,” akin to non-local denoising: channels and regions uncorrelated with the learned scene embedding are attenuated. R1-W2: Clarification of Adaptive IoU Threshold. While Eq. 6 enforces a lower bound at 0.25, the dynamic term raises the threshold for larger objects, preventing over-assignment. Moreover, our Dynamic Label Assignment (Eq. 7–8) gradually increases the weight of regression IoU, mitigating any transient confidence bias. R2-W1: Broader baselines and ablation scope. Due to space constraints, we primarily compared six leading CNN and DETR variants (Table 1), but will explore broader baselines (e.g., YOLOv7, TransUNet) and vision-language models (e.g., LLaVA) in future work. R2-W2: Generalization analysis beyond PerioXrays. We plan to evaluate PerioDet on the UFBA-UESC dataset in future work to assess cross-dataset robustness. R2-W3: IDC functional form justification. We performed a grid search over the exponent λ and the scaling coefficient to maximize small-lesion recall without degrading large-lesion precision. This form gracefully interpolates between a conservative minimum (0.25) for very small boxes and a higher threshold for larger regions, avoiding the imbalance introduced by a fixed IoU. R2-W4: Discussion on Annotation Ambiguity and Inter-Rater Variability. We thank the reviewer for this important point. We will compute inter-annotator agreement, reporting both Dice overlap and Cohen’s κ, over a random subset of 200 images. R2-W5: Clarification of error analysis. We appreciate the reviewer’s suggestion. In fact, we performed a detailed error analysis, including identifying false positives and missed micro-lesions under low contrast, and generated qualitative examples. However, due to the 8-page limit, these results were omitted from the main paper. R2-W6: Please refer to R1-W1. R3-W1: Details about dataset diversity. We will expand the dataset description to include details on imaging equipment and acquisition protocols in the paper. R3-W2: Size of human-computer study. Although six dentists and 100 images yield statistically significant improvements, we acknowledge the sample size limitation. We will add confidence intervals and effect-size analysis to strengthen this section in future work. R3-W3: Novelty of subcomponents. While attention and adaptive thresholding have appeared in other domains, BDA’s explicit combination of channel weighting with scene-anchored similarity, and IDC’s dynamic IoU threshold formula with epoch-dependent weight α are tailored innovations for panoramic radiographs. We will clarify these distinctions and cite related Transformer-based denoising and thresholding works to position our contributions more precisely.
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