List of Papers Browse by Subject Areas Author List
Abstract
Functional magnetic resonance imaging (fMRI), a noninvasive neuroimaging technique for mapping neural activity, has demonstrated substantial potential in identifying brain disease. However, clinical applications face a critical challenge: patient data are typically scarce compared to abundant healthy control samples. This severe class imbalance significantly limits the performance of classification-based diagnostic models. To address this issue, we propose the Region-Specific Anomaly Detection (RSAD) framework, which formulates the brain disease identification as an anomaly detection task. We first employ pre-training to capture normal patterns of healthy data through a reconstruction task, and then develop the discrepancy score to enhance the model’s ability to perceive potential abnormalities, thereby improving the AD performance. Specifically, we design an affinity matrix learning module and an adaptive region of interest (ROI) masking strategy to improve the performance of signal representation learning. Additionally, we propose a region-based discrepancy score weighting strategy to amplify the distinction between potential abnormalities and healthy controls by assigning higher weights to key brain regions, thereby improving the model’s ability to detect anomalies. We conduct experiments across six different brain diseases, and the superior results demonstrate that RSAD effectively enables disease diagnosis, even with extreme sample imbalances. Our code is available at https://github.com/xxx.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/1140_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/kylin1112/RSAD
Link to the Dataset(s)
N/A
BibTex
@InProceedings{SunYus_RSAD_MICCAI2025,
author = { Sun, Yusong and Chen, Dongdong and Liu, Mengjun and Shen, Zhenrong and Song, Zhiyun and Hu, Yuqi and Fei, Manman and Han, Xu and Liu, Zelin and Fang, Xingkai and Bai, Lu and Zhang, Lichi},
title = { { RSAD: Region-Specific Anomaly Detection in fMRI for Disease Diagnosis } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15975},
month = {September},
page = {485 -- 495}
}
Reviews
Review #1
- Please describe the contribution of the paper
This paper proposes a Region-Specific Anomaly Detection (RSAD) framework for fMRI-based disease diagnosis.
- 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 method and application are intreasting and descripted well.
- 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 total number of samples summed for the 6 different diseases mentioned in section 3.1 does not correspond to the 201, please reconfirm.
- In the ablation experiment,Table3 shows that the values of AP and AUC of the framework in the absence of all three parts of AML, ARoM, and Sdis are basically higher than the values of the other methods of Table1 and Table2, indicating that the choice of methods for the comparison experiment is unreasonable, and that a more appropriate and convincing AD detection method should be chosen for the comparison experiment.
- The process of selecting the hyperparameters corresponding to the variance and covariance losses in the total loss function Ltotal can be further illustrated to demonstrate the advantages of the multitasking framework. Comparison experiments can be supplemented where conditions allow to further validate the model’s contribution of the three losses to the reconstruction task under different weights (hyperparameters) (a single reconstruction loss, Lrec, can be used as one of the control groups, i.e., the hyperparameters of the latter two losses are 0).
- 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.
(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?
Both methods and experimental results are better described.
- 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
The authors propose a novel representation learning framework for disease diagnosis ( framed as an anomaly detection task ) for fMRI data.
- 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 exceptionally well-written and a pleasure to read, presenting complex concepts with clarity while maintaining technical rigor. The authors have conducted thorough baseline comparison experiments against state-of-the-art methods, providing credible validation of their approach. Particularly impressive is how the proposed method consistently outperforms existing baselines by a substantial margin across various datasets and evaluation metrics.
- 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.
Every method has inherent limitations, and these should be explicitly discussed. The authors should identify scenarios where their approach might not perform optimally or could potentially fail. Additionally, the paper would also benefit from a more in-depth discussion of potential extensions to the current method that could guide future research.
- 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
Please back these claims with citations:
- “Second, it is difficult to optimize the latent space of fMRI data simply by sequence reconstruction, due to the high noise and low information characteristic of BOLD signals.”
- “Third, existing methods typically employ reconstruction loss directly as the anomaly score, which exhibits limited sensitivity to subtle variations, particularly in conditions like brain diseases that lack distinct morphological features.”
Figure 1 caption appears to be incomplete.
- 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?
Overall, this is a strong contribution to the field. With the addition of a more thorough discussion of limitations, I recommend acceptance with minor revisions without requiring rebuttal.
- 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
In this paper, the authors propose “Region-Specific Anomaly Detection” (RSAD) method. It is an anomaly detection framework for brain disease diagnosis applied on fMRI. The brain disease is considered as an anomaly on fMRI, meaning that the pathophysiology of the disease is not learned. Instead, the patterns of “healthy” brains are learned. The method includes two keys components :
- an affinity matrix learning module to capture relationships between brain regions;
- an adaptive ROI masking for signal representation. The authors conducted experiments on UK Biobank databases. They demonstrated the effectiveness and interpretability across six brain diseases, outperforming existing methods.
- 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 clear and well written. The contribution is clearly presented and well explained.
- The authors conducted a strong evaluation including signal reconstruction and ablation study.
- The region-specific discrepancy score weighting strategy is an original approach that goes beyond global signal reconstruction and focus on local anomalies.
- The authors provided a simple but relevant method to asses the interpretability of RSAD. This helped to extract insights consistent with prior-knowledge on the brain networks correlated to anomalies linked with brain diseases.
- 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 choice of the atlas AAL-424 can be challenged through a sensitivity analysis of the chosen atlas. The impact of the atlas and the number of ROIs on the quality of the prediction could be assessed.
- The dataset size of dementia (n=10), bipolar (n=38) are limited. This makes difficult to draw strong conclusions from the benchmark on the generalizability of the findings.
- The authors did not give any order of magnitude on the computational cost. This can be helpful to compare directly to brainLM for example.
- 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?
Regardless of the positive results, the authors presented a solid research work in terms of experiment. Even though the different concepts of RSAD are known, the origintality of the work relies on the adaptation of the ROI masking and the affinity matrix for the representation learning of the BOLD signal.
- 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 reviewers and AC for the insightful comments. Below, we provide point-by-point responses to each of the reviewers’ comments. We will also make further revisions based on our responses in the final version. For Reviewer #1: 1): Inconsistency in reported sample size: Thank you for noticing this mistake. The actual number of patient samples across the six disease types is 272, not 201. We will correct this discrepancy in the revised manuscript. 2): Baselines in ablation outperform other methods, suggesting comparison may be insufficient: We agree that more fMRI-specific anomaly detection methods would strengthen the comparison. Due to the limited availability of such methods, we used state-of-the-art time-series AD models. In future work, we plan to incorporate additional unsupervised methods tailored for fMRI data. 3): Lack of details on hyperparameter selection and their contribution to performance: Thank you for the valuable suggestion. Due to space limitations, we were unable to include a comprehensive sensitivity analysis in the current version. Nevertheless, we have observed in our experiments that the variance and covariance losses play a crucial role in avoiding representation collapse and enhancing the robustness of learned features. Future work will include experiments under different weight settings, including using only the reconstruction loss. For Reviewer #2 1): Lack of discussion on method limitations: Thank you for this important point. As shown in Tables 1 and 2, our method performs relatively poorly on the Anxiety group. This may be due to the closer similarity between Anxiety and healthy control subjects in the feature space, unlike conditions with more pronounced functional abnormalities such as dementia or Parkinson’s disease. One potential direction to address this limitation is through few-shot learning, which could help enhance detection performance in subtle cases. 2): Insufficient discussion of future directions: We appreciate the suggestion. In this work, we focused on unsupervised anomaly detection using fMRI data. Future directions include expanding the dataset size and diversity, and exploring multimodal anomaly detection by integrating other neuroimaging modalities (e.g., structural MRI or EEG), which may provide complementary information and enhance performance. For Reviewer #3 1): Need for atlas sensitivity analysis: Thank you for raising this point. While AAL-424 was selected for its fine-grained regional representation, we acknowledge the importance of evaluating how the choice of atlas and the number of ROIs affect the model’s performance. We will consider conducting a sensitivity analysis in our future work to investigate the robustness of our findings across different brain parcellation schemes. 2): The dataset size for certain diseases (e.g., dementia n=10, bipolar n=38) is limited: We agree with this observation. One of the motivations for our framework is to address the challenge of small sample sizes, which are common in clinical datasets. Our approach demonstrates that meaningful anomaly detection is feasible even in such low-data regimes, where traditional classification-based approaches may fail. Nonetheless, we plan to extend the dataset with more samples and include additional disease categories in future studies to improve generalizability. 3): Missing computational cost comparison: Thank you for pointing this out. The model size of the proposed RSAD framework is approximately 26MB. In comparison, the BrainLM model used as a baseline is around 650MB—about 25 times larger. Moreover, BrainLM was trained on a dataset approximately 60 times larger than ours. This comparison highlights the computational efficiency and scalability of our method, especially in scenarios with limited data and resources.
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