List of Papers Browse by Subject Areas Author List
Abstract
Perfusion imaging guides clinical evaluation of stroke and brain tumors by characterizing tissue-level hemodynamics. Routine quantification relies on manual arterial input function (AIF) selection followed by deconvolution, producing summary maps without an executable temporal model for simulation or mechanistic insight. Tracer-dynamics-based models infer transport or compartmental parameters from perfusion time series, but do not yield clinically actionable perfusion indices (e.g., CBF, CBV, MTT) that inform diagnosis and treatment decisions. In this work, we propose HemoPIC, a physics-informed cerebral hemodynamics digital twin that explains perfusion time series through tracer mass conservation and a lumped parameter hemodynamic model. Specifically, HemoPIC solves a constrained inverse problem that jointly estimates digital twin parameters and latent states from perfusion imaging, eliminating manual AIF selection and deconvolution from routine perfusion quantification while directly producing clinically actionable perfusion summary maps. Experiments demonstrate that HemoPIC reconstructs tracer dynamics, generates physiologically consistent perfusion maps with lesion hypoperfusion patterns, satisfies central volume consistency, and yields a mechanistic hemodynamic digital twin. Code is publicly available at https://github.com/jhuldr/HemoPIC.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/2058_paper.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to the Code Repository
https://github.com/jhuldr/HemoPIC
Link to the Dataset(s)
N/A
BibTex
@InProceedings{LeeYi_HemoPIC_MICCAI2026,
author = { Lee, Yi-Chen AND Liu, Peirong},
title = { { HemoPIC: A Physics-Informed Cerebral Hemodynamics Digital Twin for Brain Perfusion } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 16894},
month = {September},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
This work introduces a new, physics-informed learning approach for perfusion imaging via lumped parameter hemodynamics modelling.
- 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 introduction of a Lumped Windkessel model proves successful in capturing patient-specific hemodynamics.
-
This is the first work, according to the authors, in capturing both inflow/outflow perfusion phases.
-
Results are in line with a conventional perfusion (deconvolution) model.
-
- 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.Lack of Comparative Baseline Analysis The current manuscript lacks a comprehensive comparative evaluation, which makes it difficult to properly contextualize the reported findings. While the authors acknowledge previous efforts in this domain, the following points must be addressed to demonstrate the state-of-the-art positioning of this work:
Direct Comparisons: No quantitative comparisons are provided against relevant prior art, specifically references [15] and [16] already cited in the text.
Physics-Informed Benchmarks: The evaluation should include comparisons with alternative modelling approaches like PINN-based ones, such as those proposed by de Vries et al. (MIDL 2024, MedIA 2023), which are highly relevant to the methodology described.
Methodological Transparency: The “single baseline” deconvolution model requires further detail. Please specify the exact approach used (e. g. , is this a block-circulant SVD approach?) to ensure the comparison is fair and reproducible.
2.Clarification of Clinical Context There is a recurring overstatement throughout the narrative suggesting that current perfusion models “do not yield clinically actionable perfusion indices (e. g. , CBF, CBV, MTT) that inform diagnosis and treatment decisions. “ This statement contradicts the extensive clinical evidence regarding EVT decision-making via perfusion mismatch, which is a standard of care in clinical routine [NEJM 2018]. I recommend the authors temper these claims and adjust the messaging to accurately reflect the established role of current perfusion software in stroke routine.
3.Impact of Diffusion Terms The authors adopt an advection-diffusion framework but opt to disregard the diffusion terms. However, prior research (e. g. , [14-16]) has demonstrated that these terms are essential for capturing physiologically relevant tracer propagation dynamics in the brain. To justify this modeling choice, the authors should provide ablation studies comparing the model’s performance with and without the inclusion of diffusion processes.
4.Evaluation of Tmax and Clinical Actionability In clinical practice, both Tmax and CBF maps are vital for defining penumbra regions and identifying perfusion mismatch. By omitting Tmax from the comparative analysis, the manuscript misses a crucial metric for determining the clinical utility of the proposed method. Including Tmax is necessary to support the authors’ claims regarding clinical actionability, especially when positioned against established clinical software.
5.Performance Metrics and Lesion Identification To better evaluate the proposed model’s efficacy in lesion identification compared to the baselines, the authors should report ROC or Precision-Recall (PR) curves. Such analysis is necessary to provide a clearer understanding of the method’s capability in lesion detection and, consequently, its potential for treatment-decision support.
de Vries, Lucas, et al. “Accelerating physics-informed neural fields for fast CT perfusion analysis in acute ischemic stroke. “ Medical Imaging with Deep Learning. 2024.
De Vries, Lucas, et al. “Spatio-temporal physics-informed learning: A novel approach to CT perfusion analysis in acute ischemic stroke. “ Medical image analysis 90 (2023): 102971.
Albers, Gregory W. , et al. “Thrombectomy for stroke at 6 to 16 hours with selection by perfusion imaging. “ New England Journal of Medicine 378.8 (2018): 708-718.
- 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.
- 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?
-
- 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
-
- 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 — marginally below the acceptance threshold, but would not mind if accepted, dependent on rebuttal
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
While the work incorporates a novel lumped parameter modeling strategy, the lack of proper baseline comparisons precludes a transparent assessment against state-of-the-art approaches. Without benchmarking against established approaches, the relative performance and actual clinical utility of the proposed model cannot be objectively assessed. Providing these comparisons is essential to contextualize the findings within this field.
- 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
This paper proposes HemoPIC, a physics-informed cerebral hemodynamics digital twin for brain perfusion imaging that aims to infer clinically actionable perfusion summary maps directly from routine perfusion data without requiring manual arterial input function (AIF) selection or deconvolution. The core idea is to couple tracer mass conservation with a regionally aggregated lumped Windkessel model, and to estimate latent inflow, outflow, volume, and resistance/compliance parameters through a constrained inverse problem. In addition to producing standard perfusion quantities such as CBF, CBV, and MTT, the method is designed to reconstruct the full tracer inflow-outflow passage and expose mechanistic latent dynamics that can support forward simulation and counterfactual analysis. This makes the paper more than a map-estimation method; it is framed as a patient-specific, physics-grounded digital twin for cerebral perfusion.
- 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’s strongest aspect is the quality of the methodological formulation. The proposed model is mathematically well grounded, clearly motivated, and very coherent for a medical imaging paper submitted to MICCAI. Rather than introducing another loosely constrained learning pipeline, the authors derive the method from tracer transport, regional inflow-outflow balance, and a lumped Windkessel closure, then formulate estimation as a constrained inverse problem with interpretable parameters and regularization. The writing is also strong: the methods section is clear and thoughtful, and the paper does a good job of explaining how the inferred latent states relate to clinically familiar perfusion concepts. On the experimental side, the paper shows several encouraging results: the produced maps appear physiologically plausible, lesion hypoperfusion patterns are visible, regional central volume consistency is reasonably preserved, and the model offers capabilities beyond standard deconvolution, including full tracer passage reconstruction and interpretable time-scale parameters. The paper is conceptually original and I love the fact that the authors have provided anonymized code implementation.
- 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 main weakness is that the empirical validation is still somewhat indirect relative to the ambition of the paper. Much of the evaluation compares HemoPIC to conventional deconvolution-derived perfusion maps on ISLES 2017 and emphasizes similarity of scale, lesion contrast, regional ratios, and central volume theorem consistency. These are useful sanity checks, but they do not fully establish that HemoPIC is clinically superior, more accurate with respect to ground-truth physiology, or more robust across acquisition settings. In several places, the paper demonstrates consistency with conventional outputs rather than a clearly stronger downstream endpoint. Specifically, the study is limited to a 44-patient ISLES 2017 cohort, which is a reasonable starting point but still small for a method making broad claims about routine perfusion quantification and digital-twin-based assessment. Another concern is that some practical parts of the pipeline remain dependent on design choices that may affect the inferred quantities, such as the construction of the shared arterial proxy curve, anatomical partitioning via segmentation and clustering, and voxelwise modulation of regional means using existing data maps. These choices are sensible, but they introduce modeling assumptions that could influence the resulting maps and deserve more discussion. While the method is presented as enabling simulation and counterfactual analysis, the paper only offers limited demonstration of these twin-specific advantages beyond reconstructed tracer curves and inferred Windkessel time scales.
- 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 has provided an anonymized link to the source code, dataset, or any other dependencies.
- 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?
I assigned this paper a Weak Accept. The strongest factor in its favor is the quality of the methodological contribution. The paper is mathematically grounded, clearly written, and proposes a genuinely interesting formulation that goes beyond standard perfusion post-processing. In particular, the coupling of tracer mass conservation with a regional lumped Windkessel model, and the estimation of latent inflow/outflow dynamics through a constrained inverse problem, is a thoughtful and well-structured contribution. I also found the framing compelling: the method is not only intended to produce clinically familiar perfusion maps such as CBF, CBV, and MTT, but also to reconstruct the full tracer passage and expose latent hemodynamic states for simulation-oriented analysis. This gives the paper conceptual originality and distinguishes it from purely deconvolution-based approaches. I did not score the paper higher because the empirical validation, while encouraging, remains somewhat indirect relative to the paper’s ambition. The experiments are mainly performed on the 44-patient ISLES 2017 cohort, and much of the evaluation emphasizes agreement with deconvolution-derived maps, physiological plausibility, lesion contrast, and central volume consistency. These analyses are useful, but they do not fully demonstrate that the method yields clearly superior clinical utility, stronger generalization, or more accurate physiological estimation beyond consistency with conventional outputs. In addition, several practical modeling choices, such as the construction of the global arterial proxy curve, the anatomical partitioning strategy, and the way regional estimates are distributed voxelwise, could meaningfully influence the final maps and deserve more discussion. I found the paper technically strong and above threshold, but the validation is not yet comprehensive enough for a stronger accept.
- 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 recommend acceptance because this paper makes a genuinely novel and well-founded methodological contribution. HemoPIC introduces a physics-informed cerebral hemodynamics digital twin that reconstructs full tracer inflow–outflow dynamics and directly produces clinically relevant perfusion maps without manual AIF selection or deconvolution. The formulation is mathematically strong, clearly presented, and substantially more original than many incremental submissions. My main reservation was that the experimental validation was somewhat indirect relative to the ambition of the work. The rebuttal addressed this reasonably well by tempering the clinical claims and adding lesion-identification analysis showing that HemoPIC preserves lesion-relevant perfusion contrast. Although further validation would still strengthen the paper, I view it as clearly above threshold overall and recommend Accept. Kudos to the authors!
Review #3
- Please describe the contribution of the paper
The authors present a method for estimating standard cerebral hemodynamic maps (CBF, CBV, MTT), solely from dynamic contrast MR and obviating the need for independent arterial input estimation. They achieve this by making physiologically informed assumptions that complement and constrain the kinetic model fits and by effectively using an image derived input function.
- 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 presented is novel because it demonstrates a newly formulated kinetic (and physical) model for a well-established system, cerebrovascular hemodynamics. -The primary benefit of the model is to eliminate the need for independent AIF estimation, utilizing physical modeling and assumptions to ensure that parameter estimation remains accurate and robust. -The mathematics are clear to follow and well presented. -The experiments presented effectively compare the proposed model to the existing standard and show it to perform comparably.
- 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.
-It would be helpful in future work to explore the assumptions made to effectively compress a complex kinetics/physics model down to five parameters. Are there regimes in which the assumptions made here don’t hold and therefore limit the application of the model?
- 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.
- 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.
(6) Strong Accept — must be accepted due to excellence
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The work presented is novel, offers a clear operational benefit to clinical users, and is well explained.
- 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 are encouraged that all reviewers recognized HemoPIC as a conceptually original, well-motivated, and mathematically grounded formulation for perfusion modeling.
We especially appreciate R2 and R3 for noting that HemoPIC is “not another loosely constrained learning pipeline”, and offers “a clear operational benefit to clinical users”.
Our central goal is to establish HemoPIC as a new foundation that simultaneously reconstructs and links observed tracer dynamics, latent hemodynamics, and clinically familiar perfusion maps — without relying on the AIF or deconvolution used in existing models.
First, we address comments on evaluation and related work.
1.Clinical claims (R1, R2)
We agree that the current manuscript demonstrates methodological feasibility, physiological consistency, and perfusion-map recovery, rather than definitive clinical superiority or treatment-decision support.
We will temper our clinical wording accordingly, and present HemoPIC as a foundation for larger-scale validation of its generalizability and clinical actionability.
2.Downstream lesion identification (R1, R2)
We agree that, beyond the direct map-consistency analyses in Figs. 2-4, downstream lesion identification provides additional evidence of clinical relevance.
We performed ROC analysis on ISLES2017 and HemoPIC-derived maps. HemoPIC matched or exceeded ISLES2017 maps in ROC AUC: 0.76 vs 0.75 for CBF, 0.58 vs 0.58 for CBV, and 0.72 vs 0.69 for MTT. These results further support that HemoPIC preserves lesion-relevant perfusion contrast.
3.Clarifications on related work (R1)
We will strengthen these in the Introduction:
(a) PIANO [15,16] is not a direct comparison: as an advection-diffusion model, its estimates are velocity (V) and diffusion (D) fields from tracer outflow, which are not standard in clinical practice. In contrast, HemoPIC is designed to connect full inflow-outflow tracer dynamics to clinical perfusion maps.
(b) ISLES2017 maps: we clarify that our deconvolution baselines were not independently tuned, but taken directly from ISLES2017, with processing details described in [21]. They serve as fixed, silver-standard clinical references, rather than hand-selected baselines.
(c) Other benchmarks, including PINN variants, are relevant but target a parallel objective: improving the speed, robustness, and accuracy of conventional AIF-based computation, whereas HemoPIC derives these maps by reconstructing the underlying hemodynamics. As also noted by R2, this makes HemoPIC fundamentally different and “more than a map-estimation method”.
Further, we clarify other points on model formulation.
4.Diffusion (R1)
The diffusion term in PIANO is not actual molecular motion as in DWI/DTI (Sec. VI-C in [15]).
Instead, it: (a) represents macroscopic intra-voxel-level tracer advection, to compensate for unresolved fine-scale flow effects at coarse perfusion-imaging voxel resolution (Sec. II-A in [15]); (b) contributes much less than advection (Sec. V-D in [15]).
In contrast, HemoPIC’s regional integration converts the diffusion term into a net boundary-flux residual rather than a voxel-wise field. We set this residual to zero as a lumped approximation, since effective inflow, outflow, and storage capture regional tracer exchange.
5.Tmax (R1)
Since HemoPIC reconstructs the full tracer passage and arterial proxy through latent hemodynamics, Tmax is naturally recoverable as a downstream measure of tracer arrival delay. Briefly, we define HemoPIC-derived Tmax as the nonnegative peak delay between the reconstructed tissue tracer curve and the inferred arterial proxy.
6.Compressed physics model (R3)
We agree that HemoPIC shows clinical consistency as a compact model, and more complex formulations may improve expressivity. These formulations may also raise identifiability challenges: we are studying when latent physiology is reliably recoverable from observations, and when more complex setups may fail.
Meta-Review
Meta-review #1
- Your recommendation
Invite for Rebuttal
- 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.
This paper proposes a physics-informed cerebral hemodynamics digital twin to estimate perfusion maps and latent dynamics without explicit AIF selection.
Reviewers agree the formulation is novel, mathematically grounded, and well motivated, but raise concerns about limited and indirect validation, missing comparisons to relevant baselines (including physics-informed approaches), and some overstated clinical claims. I suggest that the authors clarify those raised concerns.
- 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.
The reviewers generally agree that the paper presents a highly interesting, mathematically grounded, and clinically relevant formulation for cerebral perfusion modeling. The rebuttal reasonably clarifies the clinical framing, the relation to prior work, and the intended scope of the evaluation. I therefore recommend acceptance.
At the same time, there are valid concerns about benchmarking and broader validation, which I would encourage the authors to address in the camera-ready or a future journal extension.
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.
Novel, well-grounded formulation with open-source code provide. Two reviewers vote to accept including one strong accept and R1’s comparison request is a scope mismatch rather than a flaw (although could be cited as related parallel work in the discussion.
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.
This is one of the strongest papers in this batch. Reviewer #3 assigned the maximum score (6, independent of rebuttal) and Reviewer #2 upgraded to Accept post-rebuttal. The paper presents a genuinely innovative physics-informed digital twin framework for cerebral hemodynamics, and Reviewer #3’s endorsement reflects the technical depth of coupling computational fluid dynamics with neural inference. Reviewer #1’s persistent concern about clinical validation is a legitimate but appropriate limitation for a methodological contribution of this kind; it does not outweigh the majority for acceptance. The rebuttal satisfactorily addressed Reviewer #2’s concerns, leading to the upgrade.