Abstract

Gene expression profiling provides critical insights into cellular heterogeneity, biological processes and disease mechanisms. There has been an increasing interest in computational approaches that can predict gene expression directly from digitalized histopathology images. While image foundation models have shown promise in a variety of pathology downstream analysis, their performances on gene-expression prediction are still limited. Explicitly incorporating information from the transcriptomic models can help image models to address domain shift, yet the fine-tuning and alignment of foundation models can be expensive. In the work, we propose Parameter Efficient Knowledge trAnsfer (PEKA), a novel framework that leverages Block-Affine Adaptation and integrates knowledge distillation and structure alignment losses for cross-modal knowledge transfer. We evaluated PEKA for gene expression prediction using multiple spatial transcriptomics datasets (comprising 206,123 image tiles with matched gene expression profiles) that encompassed various types of tissue. PEKA achieved at least 5\% performance improvement over baseline foundation models while also outperforming alternative parameter-efficient fine-tuning strategies. We will release the code, datasets and aligned models after peer-review to facilitate broader adoption and further development for parameter efficient model alignment.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/4791_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/RunningStone/PEKA.git

Link to the Dataset(s)

N/A

BibTex

@InProceedings{PanShi_Teaching_MICCAI2025,
        author = { Pan, Shi and Chen, Jianan and Secrier, Maria},
        title = { { Teaching pathology foundation models to accurately predict gene expression with parameter efficient knowledge transfer } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
        year = {2025},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15966},
        month = {September},

}


Reviews

Review #1

  • Please describe the contribution of the paper

    This paper proposes a parameter efficient knowledge transfer work-flow for foundation model alignments, with focus on histopathology cell image and its corresponding gene expression. They use block-affine adaptation, with knowledge distillation (KD) and structure alignment (SA) losses, for facilitating cross-modal transfer. The KD loss helps domain knowledge transfer from a higher teacher model to the student model, and the SA loss preserves the structural relation in the gene expression domain space. The resulting approach result in superior performance - with proven results across four different datasets.

  • 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. The paper is easy to read and understand with respect to literature review, motivation, problem formulation and solution.
    2. It recognizes and addresses a problem in the histopathology space -with proven results across four different datasets for evaluation.
    3. The paper also studies multiple different vision models as well as parameter-efficient methods (LoRA, AdaLoRA), which is also shown as a table.
  • 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. How sensitive is PCA-Ridge regression for gene expression prediction - specifically in terms of dataset noise/quality/quantity?
  • 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.

    (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 paper is well-motivated and easy to read. It proposes a knowledge distilled framework, with parameter-efficient learning, to obtain a better efficient model.

  • 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 proposes PEKA, a parameter-efficient framework that aligns histopathology and transcriptomic foundation models using block-affine adaptation and cross-modal distillation. It achieves superior gene expression prediction across multiple spatial transcriptomics datasets.

  • 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. The paper proposes a novel and well-motivated framework, PEKA, that effectively aligns histopathology and transcriptomic foundation models using parameter-efficient strategies.

    2. PEKA achieves high performance while modifying only ~5% of model parameters, making it computationally efficient and scalable.

  • 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. The evaluation is limited to a single dataset (HEST), and generalizability would be better demonstrated with external validation on other cohorts or platforms.

    2. Although PEKA emphasizes parameter efficiency, its benefit may be less pronounced for backbones like UNI, where full fine-tuning is still computationally feasible and may yield better performance.

    3. novelty is limited

  • 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 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.

    (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?

    See above for details

  • 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

    The paper proposes PEKA (Parameter Efficient Knowledge trAnsfer), a novel framework for predicting gene expression from histopathology images by aligning image foundation models with transcriptomic models in a parameter-efficient manner. PEKA combines Block-Affine Adaptation, knowledge distillation, and structure alignment losses to transfer biologically meaningful information from transcriptomic representations into image-based models. Evaluated on over 200,000 image tiles from multiple spatial transcriptomics datasets, PEKA consistently improves gene expression prediction performance by at least 5% over existing baselines, demonstrating both effectiveness and generalizability.

  • 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. Timely and relevant problem: Predicting gene expression from pathology images is a highly impactful and emerging area that can greatly reduce reliance on expensive sequencing technologies like scRNA-seq or ST in clinical workflows.
    2. Cross-modal alignment formulation: The proposed PEKA framework explicitly addresses the domain gap between morphological and molecular modalities by integrating knowledge from transcriptomic foundation models. This modality-aware adaptation is a novel and meaningful direction.
    3. Parameter-efficient design: By leveraging Block-Affine Adaptation, the framework maintains parameter efficiency, which is essential for practical deployment and scalability on large histopathology datasets.
    4. Strong empirical results: The method is evaluated on a large and diverse spatial transcriptomics dataset, demonstrating robustness across tissue types. The consistent ≥5% performance gain over baselines and other fine-tuning strategies strengthens the claims.
    5. Clear real-world impact: The work is clinically relevant, and the proposed solution has the potential to make molecular-level insights more accessible and affordable in routine pathology.
  • 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.

    Missing comparison with full fine-tuning during knowledge transfer: The core claim of PEKA is parameter-efficient knowledge transfer, but the paper does not compare its performance against full fine-tuning of the image foundation model during the knowledge transfer stage. Including this baseline would provide a stronger justification for the efficiency-accuracy trade-off achieved by PEKA.

  • 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?

    I gave an Accept because the paper proposes a technically sound, novel, and timely framework to address an important problem in computational pathology—predicting gene expression from H&E-stained slides in a parameter-efficient and biologically informed way. The integration of transcriptomic knowledge into image foundation models is well-motivated and addresses a critical domain gap. The empirical results are strong, the potential impact is high, and the framework is both scalable and clinically relevant.

    Although the paper would benefit from additional clarity in certain methodological components and earlier code release, the core contributions are sufficiently significant to warrant acceptance.

  • 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 sincerely appreciate all reviewers’ valuable feedback and constructive suggestions.Following the suggestions of our reviewers, we will improve our method and experiments in our following work and journal paper in the near future. In PEKA, we chose to implement PCA-Ridge regression following the established evaluation protocol by Jaume, G., et al. for the HEST1K dataset to minimize experimental variables. We acknowledge that analyzing the model’s robustness to variations in data quality, noise levels, and sample quantity represents an important direction for future work. We also agree that comparison with full fine-tuning serves as a critical benchmark. We are currently conducting additional experiments examining performance across a spectrum of trainable parameter configurations, ranging from our parameter-efficient approach to full fine-tuning. This comprehensive analysis of the efficiency-accuracy trade-off will be included in our forthcoming journal submission.When preparing this paper for MICCAI2025, HEST1K was the largest publicly available benchmark for this task. We recognize the value of external validation and are expanding our evaluation to include more diverse datasets in follow-up work. Similarly, we plan to test additional backbone architectures beyond UNI to better quantify parameter efficiency benefits across different model sizes. These insightful comments have helped shape our ongoing research direction, and we are committed to addressing these points thoroughly in our future work.




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



back to top