Abstract

Alzheimer’s disease (AD) pathology begins years before clinical symptoms, making early and reliable diagnosis from a single modality inherently uncertain. Growing evidence links the gut microbiota to early AD through the gut-brain axis, but direct microbiome profiling is highly heterogeneous and temporally unstable, limiting its utility as a deployable screening biomarker. To bridge this gap, we propose a gut microbiota-informed genetic surrogate-driven multimodal framework for AD diagnosis (MiGAD), which incorporates gut-brain axis related genetic evidence and distills multimodal knowledge into an MRI-only model for deployable screening. Our framework comprises three stages. First, we perform microbiota-informed candidate variant selection by integrating multi-cohort microbiome genetic evidence and derive transferable subject-level genetic representations. Second, we perform aligned cross-modal token fusion with a hierarchical Transformer to integrate MRI, clinical tabular features, and genetic tokens for AD classification. Third, to handle missing peripheral modalities in real-world screening, we distill the multimodal teacher into an MRI-only student model. We evaluate our approach on 6,500 participants from UK Biobank and ADNI. The multimodal model achieves ACC=0.878 and AUC=0.963, while the distilled MRI-only student improves AUC from 0.825 to 0.910 over an MRI-only baseline. These results demonstrate that microbiota-informed genetic surrogates provide stable complementary signals for AD staging while preserving real-world usability.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/3393_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)

UK Biobank dataset: https://ukbiobank.dnanexus.com ADNI dataset: https://adni.loni.usc.edu/data-samples/adni-data/

BibTex

@InProceedings{LiaZhi_MiGAD_MICCAI2026,
        author = { Liang, Zhichao AND Zhao, Shilun AND Li, Fan AND Lian, Zifeng AND Zhou, Weilin AND Yan, Wei AND Lam, Chi Kin AND Tan, Tao AND Shen, Dinggang},
        title = { { MiGAD: Gut Microbiota-informed Genetic Surrogates for Multimodal Alzheimer’s Disease Diagnosis } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 16886},
        month = {September},
        page = {pending}
}


Reviews

Review #1

  • Please describe the contribution of the paper

    The paper proposes MiGAD, a novel framework for Gut Microbiota-informed Genetic Surrogates for Multimodal Alzheimer’s Disease Diagnosis, which integrates gut-brain axis mechanisms into AD diagnosis using stable genetic proxies instead of volatile direct microbiome data. The main contributions are threefold: (1)Microbiota-informed Genetic Surrogates: It introduces a pipeline using Mendelian Randomization and colocalization on GWAS data to identify 12,634 genetic variants that serve as stable, transferable proxies for gut microbiome status, overcoming the heterogeneity of direct microbiome profiling. (2)Aligned Cross-Modal Fusion: It develops a hierarchical Transformer with explicit InfoNCE contrastive alignment to robustly fuse MRI, clinical, and genetic surrogate tokens, achieving state-of-the-art performance (AUC 0.963) on 6,500 participants from UK Biobank and ADNI. (3)Deployable Knowledge Distillation: It proposes a multimodal-to-MRI distillation strategy that transfers gut-brain axis knowledge into an MRI-only student model. This significantly improves screening accuracy in resource-limited settings (AUC from 0.825 to 0.910), enhancing real-world clinical usability.

  • 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) Novel Methodological Formulation: The paper innovatively proposes “Microbiota-informed Genetic Surrogates” to capture gut-brain axis signals. By leveraging Mendelian Randomization on GWAS data, it bypasses the instability of direct microbiome profiling, using stable host genetic variants as reliable proxies for AD diagnosis. (2) Clinical Deployability via Distillation: The proposed Multimodal-to-MRI knowledge distillation strategy effectively addresses data scarcity in real-world screening. The distilled MRI-only student model significantly outperforms baseline imaging models (AUC 0.825 vs. 0.910), demonstrating strong practical utility where peripheral data is unavailable. (3) Robust Cross-Modal Alignment: The integration of an explicit InfoNCE contrastive learning objective ensures semantic consistency across heterogeneous modalities. Ablation studies confirm that this alignment mechanism is critical for maximizing the performance of the hierarchical Transformer fusion architecture. (4) Rigorous Large-Scale Validation: The method is evaluated on 6,500 participants from two distinct cohorts (UK Biobank and ADNI). This large-scale, cross-cohort validation demonstrates the model’s robustness and generalizability across diverse demographic and clinical populations.

  • 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) Fundamental Flaw in Causal Logic and Surrogate Validity: The core premise relies on the unproven assumption that selected genetic variants effectively proxy for gut microbiome composition. However, reverse causality is a significant confounder: AD pathology may influence microbiome composition via changes in diet, medication, or behavior. The paper fails to empirically validate that these variants predict actual microbial profiles in the study cohort. If this link is weak, the “microbiota-informed surrogates” are merely standard AD-risk loci, rendering the proposed mechanistic innovation illusory. (2) Modality Gap in Knowledge Distillation: Stage III forces an MRI-only student to mimic multimodal teacher representations (h_T). However, MRI captures structural changes, whereas genetic/clinical data reflect systemic states. It is questionable whether structural imaging alone can truly reconstruct these peripheral signals. The performance gain may stem from fitting the teacher’s output distribution rather than learning genuine gut-brain mechanisms, a claim unsupported by interpretability analysis linking student attention to biologically relevant regions. (3) Unfair Comparison with Multimodal Baselines: The evaluation in Table 3 is methodologically inconsistent. Only MADDi integrates three modalities (imaging, tabular, genetics), while HyperFusion and DAFT use only imaging and tabular data. Comparing MiGAD—which leverages genetic surrogates—against baselines lacking genetic input fails to isolate the specific benefit of the proposed selection strategy. A fairer assessment would require baselines that also incorporate genetic data (e. g. , PRS-enhanced models). (4) Insufficient Experimental Details: Key reproducibility details are omitted, including the dataset splitting strategy (e. g. , random or site-based) and specific MRI data augmentation techniques. (5) Poor Readability and Undefined Notations: The manuscript lacks clarity in its mathematical formulation. Several notations are introduced without explicit definitions, such as the classification loss L_cls(Eq. 6), the projection function fm(*) (Eq. 4), etc.

  • 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 provide sufficient information for reproducibility.

  • 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 proposes MiGAD, a novel framework that integrates gut-brain axis mechanisms into AD diagnosis using stable genetic proxies. Its main strengths are the innovative use of microbiota-informed genetic surrogates to bypass microbiome instability, robust performance on large cohorts (UKB/ADNI), and a practical knowledge distillation strategy that enables deployable MRI-only screening. However, the score is limited by several weaknesses: the biological validity of the genetic surrogates is not empirically validated against actual microbiome data; the comparison with multimodal baselines is inconsistent regarding modality usage; and the manuscript lacks key experimental details (e.g., data splitting) and clear mathematical definitions. Despite these issues, the methodological novelty and clinical applicability warrant 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



Review #2

  • Please describe the contribution of the paper

    This work involves a microbiota-informed genetic surrogate pipeline that includes GWAS + Mendelian randomization + colocalization to derive features for classifying Alzheimer’s disease. This work uses a multimodal Transformer with cross-modal alignment between MRI, tabular, and genetic modalities. This work then creates a knowledge distillation framework by transferring multimodal knowledge into an MRI-only deployable model. The work uses data from both the UK Biobank + ADNI (n ≈ 6,500) and demonstrates improved MRI-only performance via distillation.

  • 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 goes beyond typical MRI + tabular data by introducing biologically motivated genetic features. Specifically, using a microbiota-informed genetic surrogate approach is relatively novel. The topic is clinically relevant with a focus on early Alzheimer’s diagnosis using realistic missing-modality constraints. The distillation into MRI-only model is practical and well-motivated. The dataset is large at n=6,500.Overall, this work shows alignment with MICCAI through inclusion of multimodal medical imaging, imaging-genetics integration, and addressing clinical AI deployment challenges.

  • 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 information is broad and could be strengthened. Some components, such as the genetic pipeline are under-explained, which also limits potential reproducibility and understanding of what is actually driving the gains in performance compared to the other models. Some of the findings are overstated - such as this work capturing gut-brain axis mechanisms. There is an inherent assumption that the relationship between the microbiome and genetics is directly relevant to Alzheimer’s in a causal way, though that is not definitive. It could be made more clear how the data is pooled or split among the cohorts included.

  • 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 provide sufficient information for reproducibility.

  • 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?

    Strong technical design and integration Clinically relevant and deployable direction Large-scale evaluation Novel angle via microbiota-informed genetics

  • 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

    This paper proposes a multimodal framework for AD diagnosis that uses genetic variants linked to gut microbiota as stable surrogates for gut-brain axis signals. It fuses MRI, clinical data, and these genetic features using a hierarchically aligned Transformer, and then distills the multimodal model into an MRI-only student to enable practical deployment when non-imaging data are unavailable.

  • 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.

    S1: by distilling multimodal knowledge into an MRI-only student, this paper try to tackle a real-world problem where genetics/some other non-imaging modality are often unavailable in frontline screening. S2: The genetic surrogate approach is well-motivated. S3 The AUC improvement achieved by the distilled MRI-only student demonstrates the effectiveness of the knowledge transfer.

  • 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.

    W1: The knowledge distillation evaluation lacks comparison with simpler alternatives. W2: The clinical tabular features are not described. W3: Comparison with simpler genetic encoding baselines is missing.

  • 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 provide sufficient information for reproducibility.

  • 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?

    strengths S1-S3.

  • 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.

    Both R2 and R3 recommended accept, with R1 recommending a weak accept. All three reviewers acknowledged the methodological novelty, biological motivation, and clinical deployability of the paper, and the large-scale cross-cohort validation on 6,500 participants from UK Biobank and ADNI was also consistently recognized. The proposed method is novel, the experiments are solid, and the clinical significance is clear. Acceptance is recommended.



back to top