List of Papers Browse by Subject Areas Author List
Abstract
Human tissue samples exhibit remarkable cellular and structural diversity, where alterations in the spatial arrangement of cells can signal the onset or progression of disease. Therefore, characterizing these spatial cellular interactions and linking them to clinical endpoints is critical to advance our understanding of disease biology and improve patient care. In this work, we introduce a \emph{band descriptor} that quantifies the local neighborhood of each cell by computing the relative abundance of neighboring cell types using concentric bands. We demonstrate the efficacy of our approach by highlighting two key benefits: it enables the unsupervised discovery of \emph{spatiotypes} (substructures defined by local cellular configurations), and it provides an explicit encoding of spatial context in cell-level graphs —
capturing long-range cell interactions across tissue. Our experiments in a lung tissue cohort reveal distinct spatial patterns of cellular arrangement that differentiate control from disease samples and may also reflect disease progression (unaffected, less affected, or more affected). Furthermore, by explicitly modeling spatial context, our band descriptor enhances node-level representations, enabling an end-to-end Graph Neural Network (GNN) to achieve high accuracy in a clinical prediction task with fewer layers. This reduction in network depth decreases over-smoothing and improves interpretability, underscoring our approach’s potential for broad adoption in tissue-based studies and clinical applications.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2025/paper/4071_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/imuhdawood/BandDescriptor
Link to the Dataset(s)
N/A
BibTex
@InProceedings{DawMuh_Unsupervised_MICCAI2025,
author = { Dawood, Muhammad AND Thomas, Emily AND Cooper, Rosalin AND Pescia, Carlo AND Sozanska, Anna AND Ryou, Hosuk AND Royston, Daniel AND Rittscher, Jens},
title = { { Unsupervised Discovery of Spatiotypes and Context-Aware Graph Neural Networks for Modeling Clinical Endpoints } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2025},
year = {2025},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15970},
month = {September},
page = {603 -- 612}
}
Reviews
Review #1
- Please describe the contribution of the paper
This manuscript introduces a band descriptor to model the multi-scale spatial context of individual cells within tissue sections, using concentric bands to encode neighboring cell-type distributions. The authors apply this descriptor in to: (i) cluster cells into “spatiotypes” in an unsupervised manner, (ii) enhance node representations in graph neural networks for clinical endpoint prediction The method is evaluated on a spatial transcriptomics dataset of lung tissue (Xenium), and shows good results in both segmentation of microenvironments and GNN-based classification of fibrosis severity
- 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.
-Good Biological motivation: Understanding spatial cell arrangements is clinically important, particularly in diseases like pulmonary fibrosis -Interpretable method: The band descriptor is intuitive and biologically logic, providing a clear summary of local microenvironments. In addition, the descriptor directly supports explanation at the node -Improved prediction with GNNs: The method reduces the need for deeper networks
- 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.
- Lack of sufficient novelty: While the band descriptor is well-motivated, it closely resembles existing radial context descriptors (e.g., shape contexts, spatial histograms), and no strong comparison is made to them -No direct baselines or ablations: The paper presents good results, but doesn’t compare to alternative spatial encodings, such as simple radial density maps, GNNs using classical cell-type graphs, or standard morphological features
- The number of bands and their size (100μm–500μm) are not justified
- Single-dataset evaluation: The method is only tested on lung tissue from one spatial transcriptomics dataset. No cross-dataset validation, simulated data, or multi-organ variation is compared
- Claims of broad applicability (e.g., to 3D and other imaging modalities) are not supported by comparison -Limited discussion on limitations and generalisability: The authors make strong claims about biological insight and interpretability, but visualisations are limited and biological hypotheses are not validated beyond correlation -Reproducibility: Although the method is simple, no code or supplementary material is available to reproduce results or explore sensitivity to hyperparameters
- Please rate the clarity and organization of this paper
Poor
- 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.
- 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.
(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?
To move toward acceptance, the authors should consider the following in their rebuttal:
- Include comparisons to alternative spatial encoding methods, including simpler baselines (i.e, radial histograms or existing cell-cell graph models). -Provide a sensitivity analysis on the band descriptor parameters (number of bands, size, type encoding). -Strengthen the claim of generalisability by testing on an additional dataset -Clarify plans for code release or detail a reproducibility
- 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 introduces a method called band descriptor. The method can quantify the local neighborhood of each cell and encoding the spatial types in cell-level graphs. Then the graph neural network is leveraged to reveal the spatial patterns between each cell.
- 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 ‘band descriptor’ idea is novel in this situation. The paper sets a radius set(3 types) to describe the local neighborhood of each cell, which contains the surrounding cell information. Then the GMM model is conducted to perform clustering. The pipeline is flexible and the hypothesis made by the author is not strict, suggesting that the method can be expanded to a similar situation.
- 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.
There’s no baseline comparison to provide a direct understanding of how the method works and how efficient it is.
- 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?
NA
- 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 study is well designed and executed and timely using spatial technology from 10X genomics and applied some unique AI analytic approached to find the spatial and cellular information which might impact clinical endpoints.
- 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 timely study really try to answer a question that puzzling the entire spatial community. is the cellular components alone can be informative sufficiently or really need spatial info? The results highlight two key contributions of the proposed band descriptor. First, by encoding the spatial composition of different cell types around each cell using concentric bands, it enables the discovery of spatiotypes that differentiate disease samples from controls in an interpretable and clinically meaningful way.
- 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.
there are plenty missing info in the study which I am not sure whether it is due to the words limit.
The study is nicely done (esp methodology, initiatives etc.) but the results as well as discussion are pretty short - what a pitty.
- 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 think the study is pretty timely for the spatial tech and trying to understand the cellular features and spatial features if they are predicting clinical outcome alone or interactively.
I would like to see more studies like this presented in MICCAI as the big data set of spatial (a new sequencing era) - omics will be impacting the entire digital healthcare field in no time.
however, the study weirdly have a short results and discussion where I thought the authors can highlight the impact better and clearer.
- 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
Deciphering the spatial organization of cells — and its alteration in disease — is crucial for understanding pathobiology and informing diagnosis. Current methods cannot identify clinically relevant, recurring cellular microenvironments across multiple whole-slide images (WSIs) in a data-driven and interpretable way. The proposed band descriptor captures multi-scale spatial context of individual cells within tissue sections, enabling spatial analysis of cell-type distributions, particularly multiplexing and image-based spatial transcriptomics.
Our approach offers two key benefits: 1) unsupervised discovery of spatiotypes — substructures defined by local cellular configurations; and 2) explicit encoding of spatial context in cell-level graphs. All reviewers provided constructive comments and acknowledged the following: R2- the study is timely and interpretable, enabling clinically meaningful spatiotype discovery and addressing the question of added value of spatial information beyond cellular composition; R3- the approach is novel and flexible; R1 highlighted the approach’s interpretability, noting its ability to clearly summarize the local microenvironment, improve GNN node-level explanations and predictive performance, and reduce reliance on deeper GNN that may be prone to oversmoothing.
Response to Reviewers’ Comments
Innovation (R1) While neighborhood analysis is a known concept, to our knowledge, no prior work has systematically modelled the tissue microenvironment in this manner (acknowledged by R3). We do not assert superiority over shape context or other spatial metrics; rather, we provide a purposely interpretable, low-dimensional, and computationally efficient descriptor. It scales nearly linearly with the number of cells compared to shape context quadratic complexity, making it ideal for WSI level analysis.
Baseline/ablations (R1, R3) Existing methods do not explicitly model the microenvironment of individual cells in a structured, multi-scale manner. In response to R2’s note on missing baselines and benchmarking against cell–cell graph methods, we highlight the ‘Cell Type’ experiment (Fig. 3). Using identical graph connectivity, one-hot encoded cell types as node-level features yielded an AUROC of 0.850 ± 0.199, compared to AUROC of 1.0 ± 0 with band descriptors — demonstrating that explicitly modelling the microenvironment enhances both predictive performance and interpretability (Fig. 3, Table 1). The figure also indirectly addresses R1’s suggestion of radial-histogram baseline (GNN with one message-passing layer). In our cell–cell graph, each cell in connected to its neighbours with in a fixed radius r. The GNN aggregates the one-hot encoded type vectors of neighbouring cells, effectively producing a single-ring radial histogram of local cell-type abundance. If adjacency alone were sufficient — without the need for a multiscale descriptor — the GNN should have been able to leverage the structure encoded in adjacency matrix directly. However, this baseline performs poorly (AUROC of 0.63 ± 0.31). Deeper GNN (e.g., 9 layers; see Fig. 2) can capture such structure, but result in a loss of node-level interpretability (Table 1).
Sensitivity Analysis (R1) Due to space limitations such experiments were omitted from the original submission. While these could be included in a final submission, we regret that the submissions rule does not allow such an inclusion.
Validation on Additional Dataset (R1) We acknowledge the importance of validation on additional datasets. However, due to the substantial cost and time required to conduct additional experiments, these were not included in the current study. We intend to incorporate such validation in a future extended journal version.
Reproducibility (R1) To facilitate reproducibility, a GitHub repository containing implementation details, data source references, and supplementary material will be included in the camera-ready version.
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.
Reject
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
There are a few score-driving critiques that remained unaddressed in the rebuttal. The authors claim to have performed a sensitivity analysis yet omit it from both the manuscript and rebuttal despite having completed it beforehand, and their assertions of broad applicability and generalizability lack supporting evidence.
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’
N/A