List of Papers Browse by Subject Areas Author List
Abstract
Neurological diseases encompass a diverse range of conditions such as neurodegenerative diseases and neurodevelopmental disorders. Developing a general model to assist in the diagnosis of multiple neurological diseases is essential in clinical practice, as it can help reduce misdiagnosis rates and alleviate the burden on physicians. However, most existing diagnosis models are designed for specific neurological disease scenarios and show poor performance when applied to multiple diseases. To this end, we present a semantic-assisted framework, called Neuro-AMS, a Neuro-informed Age-aware and Medical knowledge-integrated Strategy for diagnosis of multiple brain disorders. Specifically, we employ a vision encoder based on age-aware strategy to further enhance performance by leveraging the potential relationship between age and neurological diseases. Additionally, we extract semantic features from labels and integrate corresponding medical knowledge embeddings, constructing knowledge-level label features with enhanced semantics. These knowledge-level label features guide the vision encoder for capturing higher-level semantic representations through the alignment of image-text pairs. Our method is evaluated on four public brain disease datasets, and experimental results demonstrate that our method achieves consistent and statistically significant improvement compared with three public benchmarks and three specialized models.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/3514_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
N/A
Link to the Dataset(s)
N/A
BibTex
@InProceedings{ZhaZhe_NeuroAMS_MICCAI2025,
author = { Zhang, Zhenguo and Teng, Lin and Zhao, Nan and Liu, Yuxiao and Qiu, Zhaoyu and Weng, Zehao and Kong, Jinwei and Shi, Feng and Shen, Dinggang},
title = { { Neuro-AMS: Neuro-informed Age-aware and Medical Knowledge-integrated Strategy for Diagnosis of Multiple Brain Disorders } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15974},
month = {September},
page = {422 -- 432}
}
Reviews
Review #1
- Please describe the contribution of the paper
They present a semantic-assisted framework for diagnosis of multiple brain disorders.
- 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 proposed is reasonable to me. The encoders and decoders are well-defined. Using age as a feature is great and will improve the results.
- 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.
- Not sure about the details of Table 2. Such as for AD, are you doing the experiment for AD vs others? Treating this as a binary classification problem?
- You claimed this framework is built for multiple diseases. But for the experiment it’s still only on alzheimer’s disease diagnosis. Maybe the ablation study is for multi diseases, but the details are still not clear.
- Why just use age as the related features?
- 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
- Any evidence shows that clinicians cannot deal with complex image data, according to paragraph 1?
- Aren’t there any VLMs pretrained on 3D data in medical domain before?
I will reconsider my score after rebuttal.
- 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.
(3) Weak Reject — could be 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 is not clear in some detials, but the method is reasonable to me.
- 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.
I think the author feedback is useful and has solved my problems. I believe this work is interesting and deserve an accept, although they should clarify more details about their experiments in the manuscript.
Review #2
- Please describe the contribution of the paper
This paper presents a novel VLM framework, named Neuro-AMS, which is tailored for diagnosis of multiple neurological disorders. The authors mainly introduce two key components including a text-aware encoder to integrate disease semantics and an age-aware vision encoder to leverage age information to improve the representation learning of the imaging data. The proposed model aligns age-aware imaging embeddings and knowledge-augmented label features / text embeddings in the similar fashion of CLIP. Experimental results on 4 brain disease datasets across multiple neurological diseases show the proposed method outperforms baseline VLMs such as CLIP and BiomedCLIP.
- 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 overall well-written with clear motivation and presentation for the methodology and experimental results; The topic of developing novel VLM framework by incorporating domain knowledge and age information for brain disorders is quite relevant for clinical demand as well as the research interests of the community
- The incorporation of domain knowledge from UMLS and age information is novel and serves as the major contributions for the paper; The authors have successfully demonstrated the usefulness of the proposed modules through ablation studies (Table 3.)
- The authors have utilized relatively comprehensive datasets and conducted experiments over multiple neurological diseases to show that the proposed Neuro-AMS can achieve better performances than other 3 VLM benchmarks, and obtain overall better results comparing against SOTA disease-specific models
- 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 authors claim that age could be a critical diagnostic factor for brain disorders, however, there seems to lack in depth analysis of how the age-awareness impacts different disease affects, since diseases could have different relationships with age, e.g., we know that AD and MCI could be strongly age-driven, but for ADHD and ASD, age could potentially be a confounding factor; I would suggest the authors to provide a detailed analysis for the effect of age-awareness module for each disease
- For experimental results, it seems that the proposed model does not outperform the disease-specific model especially for AD and NC, as authors could provide more explanations for these failure cases
- It would strengthen the paper if the authors could provide model interpretability for the diagnostic task such as providing saliency map and checking what regions of the MRI data mostly align with a specific textual class or prompt
- 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?
This paper is overall relevant and builds upon the recent advances of medical VLM, which could be beneficial for the community. The current presentation and results are reasonable, and there are certainly some places for further improvements.
- 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
Based on contrastive CLIP, this paper proposes a novel framework to leverage disease semantic to guide the learning of discriminative representations of vision encoder for identifying multiple neurological diseases. A text-aware encoder is designed to combine the semantic of the labels and disease-related medical knowledge into text features. The vision encoder introduces the age-related information into image features to align age and brain structure in a shared latent space. The results show that the proposed method achieves better performance compared with other methods in most diseases except for AD.
- 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 paper represents an interesting task where multiple neurological disorders are diagnosed. This task is close to the clinical scenario in real-world. The proposed framework is novelty. The labels are not considered as discrete vectors, but is incorporated into the pretrained CLIP to leverage its semantics. The text information includes not only label information but also medical domain knowledge. Age-guided features serves as diagnostic priors for disease diagnosis.
- 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 experimental evaluation did not use cross-validation. Only a part of samples are selected from the four dataset, but no rationale is provided. The proposed model shows the overly optimistic results, without any interpretable reasons. The significance test is not provided.
- 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?
It is generally easy to read, has a good structure, and clearly explains the key points of the study. This paper provides new insights into the diagnosis of neurological disorders.
- 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 AC and all reviewers for their insightful comments. Below are our point-by-point responses.
- Binary or multi-class classification problem (R2) We clarify that our proposed Neuro-AMS framework performs a unified 6-class classification rather than decomposing the task into multiple binary classifiers. The corresponding confusion matrix is shown in Fig. 4. Table 2 presents binary classification results for LEFM_s (ADHD vs. NC) and DCNN (ASD vs. NC), a 4-class task (AD vs. EMCI vs. LMCI vs. NC) for Hierarchical AD, and our proposed Neuro-AMS framework is evaluated in a more challenging multi-disease scenario (AD vs. EMCI vs. LMCI vs. ASD vs. ADHD vs. NC), demonstrating its capacity for joint diagnosis. Ablation studies in Table 3 are also conducted under the same 6-class classification setting.
- Detailed analysis of results (R2 & R3) As shown in Table 2, our method achieves notable improvements on ASD (+10.84%), ADHD (+36.57%), EMCI (+13.11%), and LMCI (+15.19%) compared to disease-specific baselines, with slight performance drops on AD (−3.2%) and NC (−6.08%). The minor drops on AD and NC may be a trade-off when optimizing for multi-disease classification. First, our model tackles a more challenging 6-class classification task, while the disease-specific methods are designed for binary or 4-class tasks, which are easier. Second, despite this increased difficulty, our model achieves the best overall performance (79.24%) across all categories. We will further discuss this point in the final version.
- Model interpretability & age prior (R1 & R2 & R3) Our framework incorporates age as a domain-guided prior, as it plays a critical role in neurological diseases. The age-awareness module modulates feature learning based on age, enabling the model to learn meaningful interactions rather than relying on age as a dominant factor. Ablation studies presented validate the positive impact of incorporating age-related information into our model. Additionally, the use of age in our current framework is intended as an initial demonstration of how clinically relevant demographic priors can guide the feature learning process. Other factors, such as gender and cognitive tests, also hold great potential and could be explored in future work. To address concerns about interpretability, we plan to integrate saliency-based visualizations (e.g., Grad-CAM) to summarize image regions associated with textual prompts. We also aim to provide disease-specific analyses of how age-awareness affects classification outcomes in future work.
- Fixed train-test split or cross-validation (R1) Given the dataset size (2,833 MRIs), we adopted a fixed train-test split and stratified sampling to ensure stable feature distribution while avoiding the high computational cost of cross-validation.
- Sample selection (R1) We did not include all available data from each source. Instead, we adopted a balanced sampling strategy to avoid class dominance. Specifically, for low-prevalence diseases like ADHD, we used nearly all available cases (280/285). For high-prevalence diseases like AD, we randomly selected 540 cases, roughly twice the ADHD count for balance. For intermediate-sized categories such as ASD, EMCI and LMCI, we used all available cases. NC samples were proportionally drawn from multiple datasets (see Section 3.1) to ensure fairness and reduce bias.
- Clinician interpretation (R2) Our intent was to emphasize that certain brain disorders, especially neurodevelopmental and early-stage neurodegenerative diseases, lack overt imaging biomarkers. Diagnosing such conditions from MRI requires extensive expertise and subtle judgment. Therefore, our model is designed to assist in precisely these challenging scenarios. We will revise the wording in the final version of the paper.
- Existing 3D VLMs (R2) Most existing 3D VLMs in the medical domain are based on CT and radiology reports, are not publicly available, and none focus on neurological diseases.
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’
The paper proposes a novel VLM framework for neurological disorder diagnosis, with two key components: a text-aware encoder to integrate disease semantics and an age-aware vision encoder to leverage age information. As currently medical VLM is primarily in 2D, the proposed technique for 3D brain MRI and neurological disease represents a nice contribution to the community, with promising results.