Paper Info Reviews Author Feedback Meta-Review Back to top
List of Papers Browse by Subject Areas Author List

Abstract

Intra-operative endovascular treatment of intracranial aneurysms calls for rapid and reliable hemodynamic quantification. Angiographic Parametric Imaging (API) is derived from DSA time-density curves (TDCs), yet irregular frame rates and dose-reduction gaps often make conventional pixel-wise Gamma-variate fitting unstable and computationally expensive. We propose PIC-Former, a physics-inspired continuous transformer that reconstructs a physiologically plausible TDC function ĉ(t) from irregular DSA samples. PIC-Former uses Δt-aware causal attention for inter-frame gaps and physics-inspired regularization to encourage valid contrast arrival, peak, and washout dynamics. On 52 paired DSA-4D Flow MRI cases and an independent 11-case CFD cohort, QDSA velocities computed from the reconstructed TDCs achieve relative MAEs of 3.1% and 2.01%, respectively, outperforming Gamma-variate fitting (11.8% on DSA-4D Flow MRI). PIC-Former generates full-field parametric maps for a complete DSA run in 12.45 s and remains stable under a 20% frame-drop stress test. In a retrospective intra-operative risk stratification study, the proposed pipeline improves stratification accuracy from 87.5% to 95.0% compared to the Gamma-variate-based baseline. The code and reproducibility test set is available at https://github.com/LY-SUSTech/PIC-Former.git.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026/paper/2583_paper.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to the Code Repository

https://github.com/LY-SUSTech/PIC-Former.git

Link to the Dataset(s)

https://github.com/LY-SUSTech/PIC-Former.git

BibTex

@InProceedings{LiuYan_PhysicsInspired_MICCAI2026,
        author = { Liu, Yang AND Liu, Zehua AND Liao, Xiangyun AND Duan, Chuanzhi AND Si, Weixin},
        title = { { Physics-Inspired Continuous Transformer for Fast QDSA Reconstruction } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 16888},
        month = {September},
        page = {pending}
}

Reviews

Review #1

  • Please describe the contribution of the paper

    A method is proposed to compute a continuous time density curve for any pixel in a DSA image sequence. The method relies on a continuous transformer with constraints for the output curve to be physically plausible as a time-density curve (TDC). The validity of the resulting TDC is evaluated in two important and timely clinical scenarios: perfusion measurements (quantitative DSA), and risk stratification following an intracranial treatment. The method is compared against 9 other methods, demonstrating improved performance against references such as 4D Flow MRI, or CFD computations.

  • 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.
    • original method, probably efficient, though its assessment is difficulty
    • evaluation against two important and timely clinical questions : quantitative DSA (also mentioned as perfusion DSA in the literature) ; and risk stratification for complications following intracranial aneurysms treatment
    • an ablation study is provided
    • comparison with 9 other methods, a rather large dataset is used, rather solid references were used (4D Flow MRI, with caveats below, CFD computations, actual post-treatment complications)
  • 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 method is not described in enough details, which makes its assessment very hard. It cannot be reproduced based on the current article in my opinion.
    • no code or data are provided for reproducibility, and details
    • conditions on the data are uncommon in clinical practice in my experience: irregular frame rate are rare (but the method can probably adapt regular frame rates), but more importantly the authors assume that 20 images are available in the sequence, for only the arterial phase, which is unpractical in my opinion and experience.
    • the dataset should be better described. In the current text, it appears that only internal carotid arteries were considered, with aneurysms on large vessels, which limits the applicability of the method. Using a public dataset, or making the authors’ public, would help reproducing and assessing the method.
  • 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.

  • 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

    Method:

    • appears primarily as a DSA frame interpolation technique. Haouchine et al, 2021 and Xu et al, 2024  should be discussed in that perspective
    • quantitative DSA is only applied here on the arterial phase of the injection, but other works (such as Wu et al 2025, that is cited) aim at processing the full blood flow cycle. In this perspective, the term perfusion DSA is often coined, and a number of prior works could be cited: Kosior et al 2019, Su et al 2025, Su et al 2023, Scalzo et al 2016 (prior work to Wu et al 2025)
    • Fig 1 is unclear: what does the « eye » mean? What is the meaning of the addition sign beside it, with only one input and one output (where is the second input to the addition?) ; there is only a physics loss that is mentioned for PIC-former, while the text combines it with a reconstruction loss ; the input to PIC-former is a discretized time-density curve, with potentially irregular samples, but the output is unclear: one c(t) value? Discretized curve (as is displayed)?; details are lacking on the multiscale temporal filter, that acts as preprocessing before the continuous transformer; what is the meaning of the arrow on the right of Physics Constraint Module? These are all loss terms that are computed in parallel a priori.
    • in the absence of code, more details should be given in the text: how is the output processed (computation of a regular spaced TDC? what space increment was used?); details on the multi-scale temporal filter are missing; details or reference for the kind of learnable kernel used to compute phi; provide formulas or cite references for the physics loss terms; why is interpolation used for $\hat{c}_d(t_k+\tau)$ in eq. 5 when PIC-former provides a continuous curve?
    • the input is a pixel TDC. But, during training, all pixels were probably not used as input since the majority of pixels lie outside arteries. It is unclear which pixel TDCs were selected as input to training the proposed model.

    Data and evaluation:

    • irregular frame rate for DSA is uncommon. More than 20 frame for the arterial phase, means a very high average frame rate that is not, or rarely used in my experience. The highest regular frame rate I have encountered is 6 frames per second, and many practitioners are reluctant to exposing their patients to such a high dose. With this frame rate, the number of frames for the arterial phase is at most 10, half of what is used in the paper.
    • Information is missing: frame rate interval (and average frame rate)?
    • Injection parameters are also uncommon to my experience: 5mL at 9mL per second means half a second injection. This can only, but probably hardly, be done manually, and thereafter is not very repeatable.
    • intra-arterial injection of contrast medium displays a different flow pattern from natural flow (as measured in 4D flow MRI). The paper relies on the assumption that both DSA and MRI imaging display the same flow patterns, which is not exactly true and should be discussed.
    • 4D flow MRI is very noisy, and only valid for rather large vessels, which limits its validity in the brain. Again, this limitation should be discussed.
    • 52 aneurysm cases were used for the evaluation, and 40 other cases for risk stratification. Since, in my current understanding, the method appears more suited to rather large blood vessels, it would be important to have information on the locations of these aneurysms.
    • in my opinion, one message of table 1 is that the Gamma-variate model is unfit for DSA imaging (which makes sense, since these images are provided after photometric transforms are applied (LUT)).

    • Measures of dispersion (standard deviation, confidence intervals. . . ) should be provided for the results in Table 1.p-values would also be appreciated in the analysis. MAE are point-wise reconstruction errors: how many grey levels were in the original DSA images?

    • code, and possibly datasets, should be made publicly available.

    References : Nazim Haouchine et al. , “Estimation of High Framerate Digital Subtraction Angiography Sequences at Low Radiation Dose,” in Medical Image Computing and Computer Assisted Intervention – MICCAI 2021, vol. 12906, ed. Marleen De Bruijne et al. , Lecture Notes in Computer Science (Cham: Springer International Publishing, 2021), 171–80, https://doi. org/10.1007/978-3-030-87231-1_17.

    Jayme C. Kosior et al. , “Exploring Reperfusion Following Endovascular Thrombectomy: A Proof of Concept Using Perfusion Angiography,” Stroke 50, no. 9 (September 2019): 2389–95, https://doi. org/10.1161/STROKEAHA. 119.025537.

    Ruisheng Su et al. , “perfDSA: Automatic Perfusion Imaging in Cerebral Digital Subtraction Angiography,” International Journal of Computer Assisted Radiology and Surgery 20, no. 6 (2025): 1195–203, https://doi. org/10.1007/s11548-025-03359-4.

    Fabien Scalzo and David S. Liebeskind, “Perfusion Angiography in Acute Ischemic Stroke,” Computational and Mathematical Methods in Medicine 2016 (2016): 1–14, https://doi. org/10.1155/2016/2478324.

    Ruisheng Su et al. , “Towards Quantitative Digital Subtraction Perfusion Angiography: An Animal Study,” Medical Physics 50, no. 7 (July 2023): 4055–66, https://doi. org/10.1002/mp. 16473.

    Ziyang Xu et al. , “MoSt-DSA: Modeling Motion and Structural Interactions for Direct Multi-Frame Interpolation in DSA Images,” arXiv:2407.07078, preprint, arXiv, July 9, 2024, https://doi. org/10.48550/arXiv. 2407.07078.

  • 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?

    The description of the method needs to be much improved. If possible code should be made publicly available. Measures of dispersion, and p-values, should be provided with the results.

  • 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.

    The authors have addressed my comments and I feel more confident about the value of the proposed work. Besides the improvements that the authors have committed to make, it is also important they clearly describe the procedure to extract TDCs from input DSA.



Review #2

  • Please describe the contribution of the paper

    This paper proposes PIC-Former, a physics-inspired continuous transformer for reconstructing continuous time-density curves (TDCs) from irregularly sampled DSA sequences in intracranial aneurysm interventions. The method combines delta-t-aware causal attention with physics-inspired regularization to encourage physiologically plausible contrast arrival, peak, and washout behavior, with the goal of improving downstream angiographic parametric imaging (API) and quantitative DSA (QDSA) velocity estimation. The paper further evaluates the reconstructed hemodynamics against paired 4D Flow MRI and an independent CFD cohort, and uses API-derived features for retrospective intra-operative risk stratification. Overall, the intended contribution is a fast and clinically oriented framework for making DSA-based hemodynamic quantification more robust to irregular temporal sampling and missing frames.

  • 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 addresses a clinically relevant problem and is motivated by a genuine practical need: obtaining hemodynamic information from standard intra-operative DSA without relying on computationally expensive or impractical alternatives such as CFD or 4D Flow MRI. The proposed method is well matched to the problem. In particular, the use of timestamp-conditioned / delta-t-aware attention for irregular sampling is a sensible and technically meaningful design choice, and the inclusion of physics-inspired constraints gives the framework more interpretability than a purely black-box temporal regressor. The empirical evaluation is also a strength. The authors validate the method on multiple cohorts, including paired DSA-4D Flow MRI cases, an independent CFD cohort, and a retrospective clinical risk-stratification cohort, which is broader than what is often seen in this type of work. Quantitatively, PIC-Former improves the reported hemodynamic consistency metrics over classical fitting methods and prior sequence models, remains robust under the Drop-20% stress test, and offers practical runtime for full-field map generation. The ablations are also useful in showing that both the physics constraints and the delta-t modeling contribute to the final performance.

  • 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 paper does not clearly explain how raw DSA runs are converted into reliable pixel-level or ROI-level time-density curves for API generation. In practice, DSA sequences are often affected by patient motion, table movement, residual subtraction artifacts, inconsistent acquisition timing, and other sources of temporal instability. Yet the manuscript presents the input as if per-pixel temporal signals are directly available, without adequately describing whether intra-run registration, motion correction, frame exclusion, ROI tracking, or other preprocessing steps were applied. This is an important omission because the feasibility of API depends heavily on the validity of these underlying temporal signals. A second related concern is cohort curation and representativeness. The paper appears to rely on a selected subset of studies satisfying conditions such as standard injection protocol, complete bolus passage, geometric calibration, and sufficient effective frames. These choices are understandable for method development, but the paper does not quantify how many studies were excluded, whether motion-corrupted or otherwise challenging runs were removed, or how representative the retained cases are of routine intra-procedural DSA. As a result, the reported performance may reflect a curated evaluation setting rather than realistic deployment conditions. This weakens both reproducibility and external validity, especially given the paper’s emphasis on intra-operative practicality. While the cross-modal validation is promising, the study remains retrospective and relatively modest in size, so claims about clinical readiness should be phrased cautiously.

  • 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?

    I assigned this paper a Weak Accept. The main reason is that the paper addresses a clinically relevant and technically meaningful problem: robust hemodynamic quantification from irregularly sampled intra-operative DSA. The proposed method is well motivated, and the combination of Δt-aware attention with physics-inspired regularization is a sensible contribution rather than a purely generic architectural substitution. I also found the empirical validation stronger than average for this type of work. The authors evaluate against paired DSA-4D Flow MRI, an independent CFD cohort, and a separate retrospective risk-stratification cohort, and they report improvements in the most clinically relevant downstream quantities, namely QDSA velocity estimation and risk stratification, while also achieving practical runtime. The ablation studies further support that both the Δt modeling and the physics-inspired constraints contribute to the observed performance. I did not score the paper higher because an important practical part of the pipeline remains insufficiently described. The manuscript assumes access to reliable pixel-level or ROI-level DSA time-density curves for API generation, but does not clearly explain how these temporal signals are obtained from raw DSA in the presence of motion, subtraction artifacts, and acquisition variability. Relatedly, the evaluation appears to rely on a curated subset of studies with standard injection protocol, complete bolus passage, geometric calibration, and sufficient effective frames, but the paper does not quantify exclusions or discuss how representative this subset is of routine intra-procedural DSA. This weakens confidence in real-world generalization and reproducibility. I found the method promising and above threshold, but these missing practical details make me more comfortable with a weak accept than a confident one.

  • Reviewer confidence

    Very confident (4)

  • [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.

    The paper is validated on a curated and enriched subset of DSA acquisitions obtained under a relatively standardized protocol, with exclusion of motion-corrupted, incomplete, or poorly calibrated runs. This limits external validity and means the results should not be interpreted as evidence of robustness under routine real-world intra-procedural variability. However, within this controlled evaluation setting, the method is technically strong and empirically convincing enough to merit acceptance.



Review #3

  • Please describe the contribution of the paper

    This paper introduces PIC-Former, a novel physics-inspired continuous transformer for reconstructing physiologically plausible time–density curves (TDCs) from irregularly sampled DSA data. The key contribution lies in combining Δt-aware causal attention with physics-inspired regularization to explicitly model temporal gaps and enforce valid contrast dynamics (arrival, peak, washout). This enables accurate and robust Angiographic Parametric Imaging (API) and quantitative DSA (QDSA) velocity estimation. The method demonstrates strong cross-modal consistency with 4D Flow MRI and CFD, while achieving fast inference suitable for intra-operative use and improving risk stratification performance.

  • 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.

    Novel temporal modeling for irregular medical data: The proposed Δt-aware continuous transformer explicitly models irregular sampling in DSA, which is a key practical challenge. This is particularly interesting as it directly addresses frame-rate variability and missing data, rather than assuming uniform temporal spacing. Effective integration of physics constraints: The inclusion of physics-inspired regularization (arrival, peak, washout dynamics) ensures physiologically plausible reconstructions. This improves robustness and avoids unrealistic curve fitting, which is a common issue in purely data-driven approaches. Strong clinical relevance and intra-operative applicability: The method targets real-time hemodynamic assessment during interventions, addressing an important unmet need where existing modalities (e.g., MRI, CFD) are impractical. Comprehensive and multi-level evaluation: The paper validates the approach across multiple datasets (DSA–4D Flow MRI, CFD, and clinical outcomes), demonstrating consistency across modalities and strong real-world applicability.

  • 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.

    Limited dataset size and lack of multi-center validation: Although multiple cohorts are used, the overall sample sizes remain relatively small (e.g., 52 paired cases, 11 CFD cases, 40 outcome cases), and all data appear to come from limited sources. This raises concerns about generalizability across institutions, scanners, and acquisition protocols. Incremental novelty over existing temporal and physics-informed models: While the combination of Δt-aware attention and physics-inspired constraints is well designed, similar ideas have been explored in continuous-time models and physics-informed neural networks. The novelty mainly lies in the integration and application to DSA rather than a fundamentally new modeling paradigm. Pipeline complexity and potential reproducibility issues The method involves multiple components (transformer, physics constraints, API mapping, SLS velocity estimation), which may complicate implementation and limit reproducibility without strong code and protocol details Limited analysis of failure cases: While robustness to frame drop is demonstrated, the paper lacks a detailed analysis of cases where the method fails or performs poorly, particularly in challenging anatomical or low-signal scenarios. Interpretability claims could be stronger:Although the model enforces physiological constraints, the interpretability of reconstructed TDCs and derived parameters is not deeply validated with clinical experts or quantitative physiological benchmarks.

  • 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 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?

    This paper presents a well-motivated and technically strong contribution to hemodynamic assessment from DSA. The proposed PIC-Former introduces an effective solution for reconstructing physiologically plausible time–density curves under irregular sampling, combining Δt-aware attention with physics-inspired constraints. This formulation directly addresses key limitations of existing approaches, particularly instability under sparse or non-uniform temporal sampling.

    The experimental evaluation is comprehensive and convincing, including cross-modal validation against 4D Flow MRI and CFD, as well as a clinically relevant risk stratification study. The method demonstrates clear improvements in both accuracy and robustness, while maintaining fast inference suitable for intra-operative use. The robustness to frame-drop scenarios further strengthens its practical relevance. Overall, the paper combines methodological novelty, strong validation, and clear clinical impact. Despite some limitations regarding dataset size and retrospective validation, the contribution is solid and well demonstrated, making it suitable for acceptance.

  • Reviewer confidence

    Somewhat confident (2)

  • [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 appreciate the reviewers’ constructive feedback and positive recognition of our work as ‘well-motivated’ with a ‘clinically relevant problem’ (R2), ‘novel temporal modeling’ and ‘comprehensive multi-level evaluation’ (R3), and ‘original method’ (R1). Below we address the main concerns.

Common Questions: Code & Dataset. Upon acceptance, we will release the code and test dataset. Besides, data collection is ongoing, with expanded data sharing being scheduled for journal paper.

(R1 W1,11 & R2 W1,2 & R3 W1) PIC-Former is not applied directly to any raw DSA videos. Inclusion required standard power-injector injection (5 mL at 9 mL/s), complete bolus passage, and geometric calibration. Runs/frames with severe motion artifacts, bolus contamination, incomplete passage, missing calibration, or <10 effective frames were excluded. Timestamped pixel-wise TDCs were then extracted from the full DSA sequence as model input. The final DSA database includes 40 patient/outcome cases, 391 DICOM sequences, 23F/17M, age36-71(mean52.2), covering no-complication, ischemic, and hemorrhagic outcomes, with a mean frame rate of 5.62fps(4/6/8fps), and corresponding 52 DSA–4D Flow MRI and 11 CFD cases. These preprocessing steps, inclusion/exclusion, and cohort details will be added to Sec. 3.1 in revised version.

(R1 W5-8 & R3 W3) We apologize that Fig. 1 and the related method details were not clear. The input is timestamped TDC samples and the output is ĉ(t), sampled for API/SLS. The eye denotes Δt-aware attention; ‘+’ denotes intensity/time fusion. The physics arrow indicates constraints jointly enter the loss. PIC-Former reconstructs TDCs for hemodynamic measurement rather than synthesizing DSA frames. This distinction and related perfusion-DSA references will be added in the revision.

(R1 W16) These statistics were computed but omitted due to space constraints: 4D Flow MAE was 3.1% ± 0.8% (95% CI [2.88%, 3.32%]); CFD MAE was 2.01%±0.52% (95% CI [1.66, 2.36%]). Paired tests show superiority over Gamma-variate (p<0.001) and PI-Transformer (p=0.018). These statistics have been added to Table 1.The original DICOM is 12-bit; TDCs are read from 8-bit DSA.

(R1 W3,10,12,13) The “≥20 effective frames” are not pure arterial-phase frames but frames before venous-sinus filling, a physician-confirmed, easily identifiable endpoint covering arterial arrival and early capillary passage. As quantitative DSA literature suggests, proximal injection perturbation rapidly decays downstream; thus SLS estimates vessel-level transport speed, supported by Fig. 2 (r=0.976; bias 0.57 cm/s) and CFD.

(R3 W2) PIC-Former is inspired by continuous-time modeling and PINNs, but its design is intrinsically different. Our method does not model generic continuous dynamics or impose PDE residuals. Instead, it targets DSA-TDC reconstruction by using Δt-aware attention for irregular DSA sampling and contrast-kinetic constraints: non-negativity, delayed arrival, peak/washout, and area stability. Ablations and cross-modal validation show its effectiveness for robust API/SLS-QDSA measurement.

(R3 W4,5) In Fig. 4, the misclassified Cases 35/37 are both no-complication cases predicted as higher risk—a specificity limitation. Both were caused by delayed vessel-wall injury after intervention.

(R1 W4,14,15 & R3 W1) ICA aneurysms are common and clinically important; thus, we used this homogeneous cohort for reliable validation with sufficient comparable follow-up cases, and will further expand when sufficient paired follow-up data become available. PIC-Former derives TDC-based hemodynamic measurements from routine intra-operative DSA without patient transfer or extra modality/acquisition, potentially saving intra-operative time and providing quantitative indicators within the treatment window. Future work includes multi-center generalization and extension to other cerebrovascular interventions such as AVM.



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-inspired continuous transformer to reconstruct physiologically plausible time-density curves from irregularly sampled DSA sequences for quantitative hemodynamic analysis.

    Reviewers agree the problem is clinically relevant with a well-motivated approach, but raise concerns about insufficient methodological clarity, missing details on preprocessing and data curation, and unclear assumptions regarding acquisition conditions and practical applicability. I suggest that the authors address these raised concerns in the rebuttal.

  • 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.

    There is consensus among the reviewers that the paper addresses a clinically relevant problem and proposes a well-motivated physics-inspired method. The rebuttal helped clarify the main concerns, and I therefore recommend acceptance.

    For the benefit of the research community and to facilitate reproducibility of the described results, I strongly encourage the authors to make the trained model weights publicly available as well.



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.

    This submission presents PIC-Former, a Δt-aware continuous transformer that reconstructs time–density curves (TDCs) from irregularly sampled DSA sequences using physics-inspired constraints. The reconstructed TDCs are evaluated across clinically relevant quantitative hemodynamic analysis tasks.

    All reviewers recommend acceptance after rebuttal, though Reviewer 3—who had already recommended acceptance pre-rebuttal—did not submit a post-rebuttal update. The reviews agree that the paper addresses a clinically relevant problem and provides strong experimental evaluation against multiple baselines and modalities, including 4D Flow MRI, CFD simulations, and a retrospective clinical outcome cohort. Remaining concerns focus on insufficient methodological detail, particularly regarding how TDCs are derived from raw DSA and the preprocessing pipeline, as well as questions about dataset curation and representativeness. These concerns were not fully resolved after rebuttal, but the reviewers nevertheless converged on a final acceptance decision.



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.

    Important concerns raised by reviewers (e.g., missing details on data curation / preparation, robustness / statistical analysis of the results). The paper remains limited by requiring clean, fairly long sequences, however, reviewers agree that the paper passes the threshold for acceptance given its technical strength and overall evaluation.



back to top