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

Abstract

Photon-counting computed tomography (PCCT) based on photon-counting detectors (PCDs) stands out as a cutting-edge CT technology, offering enhanced spatial resolution, reduced radiation dose, and advanced material decomposition capabilities. Despite its recognized advantages, challenges arise from real-world phenomena such as PCD charge-sharing effects, application-specific integrated circuit (ASIC) pile-up, and spectrum shift, introducing a disparity between actual physical effects and the assumptions made in ideal physics models. This misalignment can lead to substantial errors during image reconstruction processes, particularly in material decomposition. In this paper, we introduce a novel detector physics and ASIC model-guided deep learning system model tailored for PCCT. This model adeptly captures the comprehensive response of the PCCT system, encompassing both detector and ASIC responses. We present experimental results demonstrating the model’s exceptional accuracy and robustness. Key advancements include reduced calibration errors, enhanced quality in material decomposition imaging, and improved quantitative consistency. This model represents a significant stride in bridging the gap between theoretical assumptions and practical complexities of PCCT, paving the way for more precise and reliable medical imaging.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/3032_paper.pdf

SharedIt Link: pending

SpringerLink (DOI): pending

Supplementary Material: N/A

Link to the Code Repository

N/A

Link to the Dataset(s)

N/A

BibTex

@InProceedings{Yu_Material_MICCAI2024,
        author = { Yu, Xiaopeng and Wu, Qianyu and Qin, Wenhui and Zhong, Tao and Su, Mengqing and Ma, Jinglu and Zhang, Yikun and Ji, Xu and Quan, Guotao and Chen, Yang and Du, Yanfeng and Lai, Xiaochun},
        title = { { Material Decomposition in Photon-Counting CT: A Deep Learning Approach Driven by Detector Physics and ASIC Modeling } },
        booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
        year = {2024},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 15007},
        month = {October},
        page = {pending}
}

Reviews

Review #1

  • Please describe the contribution of the paper

    This paper considered the photon-counting computed tomography imaging problem, and proposed an application-specific integrated circuit model-guided model by integrating physics knowledge and using deep learning. Two networks are used for detector modeling and circuit modeling, respectively.

  • Please list the main strengths of the paper; you should write about 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 is well written, and the results look very promising. Images produced from the proposed approach indeed have better quality than the baselines.

  • Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.

    The technical aspects are not very clear to me (probably due to the lack of necessary background knowledge). See below some detailed feedbacks.

  • Please rate the clarity and organization of this paper

    Very 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 mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.

  • Do you have any additional comments regarding the paper’s reproducibility?

    N/A

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html

    • In figure 1 and section 2, the loss functions for training of detector net and the circuit net are not well explained. Any particular reason for using maximum log-Poisson loss instead of maximum log-Gaussian loss? The design of the outputs from the detector net, and the outputs from the circuit net should be better explained? Also how are the Poisson distributions are learned? Do the authors use re-parameterization approaches?

    • In figure 3, the quality produced from the proposed approach outperforms the baselines, however the consistency does not show significant improve (very similar standard deviation).

  • 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

    Weak Accept — could be accepted, dependent on rebuttal (4)

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
    • The idea is novel
    • The empirical results show the effectiveness of the proposed approach
    • Some key parts are not well explained
  • Reviewer confidence

    Somewhat confident (2)

  • [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

    N/A

  • [Post rebuttal] Please justify your decision

    N/A



Review #2

  • Please describe the contribution of the paper

    This paper describes a deep learning-based method for estimating the photon counting counts by taking the detector response and ASIC response into consideration. The material depths can also be achieved by reverse-engineering this process. Experiments show the effectiveness of the method.

  • Please list the main strengths of the paper; you should write about 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 proposed method is equipped with profound expertise in physics. The modeling of the detector response and the ASIC response is very delicate and comprehensive.
    2. The experiments are designed with fairness and can support the authors’ arguments.
    3. The effectiveness is significant judging from the experimental results.
  • Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.
    1. One of the most important contribution of the paper would be the Monte Carlo program for generating the simulation data for training. The authors did not provide the information on if they will open source it. At the same time, authors indicate that the program is very dedicate and taking many physical factors into consideration. The community will not be directly helped without open source or more details on the Monte Carlo model.
    2. For the detector domain material decomposition method, it is important to estimate the original spectrum S0(E). The authors may add some information on that.
    3. The authors should provide details on how to solve the optimization problem of Equation (8).
    4. The main diagram in Figure 1 is not that clear. Put some annotations may help.
  • Please rate the clarity and organization of this paper

    Excellent

  • 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.

  • Do you have any additional comments regarding the paper’s reproducibility?

    More details or open source of the Monte Carlo program would improve the reproducibility of the method.

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html
    1. For the detector domain material decomposition method, it is important to estimate the original spectrum S0(E). The authors may add some information on that.
    2. The authors should provide details on how to solve the optimization problem of Equation (8) because it is not trivial.
    3. The main diagram in Figure 1 is not that clear. Put some annotations may help.
  • 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

    Weak Accept — could be accepted, dependent on rebuttal (4)

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    Knowledge on physics of photon-counting detector and material decomposition. The practicality of the work on improving photon-counting CT material decomposition. Strong academic writing skills and clear organization of the paper.

  • Reviewer confidence

    Very confident (4)

  • [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

    N/A

  • [Post rebuttal] Please justify your decision

    N/A



Review #3

  • Please describe the contribution of the paper

    The authors propose a physics-guided material decomposition model for PCCT, with the aim to exploit deep learning for characterizing detector/ASIC response, incorporating physics parameters. They also show that it improves the quality of the CT reconstructed image, comparing their solution with other raw PCCT counts for which FBP algorithm has been applied.

  • Please list the main strengths of the paper; you should write about 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.

    It is a novel idea and method, which showed positive results

  • Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.

    Although the paper demonstrates the potential of such a solution through the material decomposition, it is lacking in several aspects. It is not clear how the physical parameters such as the ASIC dead time have been has been estimated for the input of the proposed network. In addition, the proposed work (framework) is not reproducible. for further research. It would be good if the authors could provide additional information on its implementation and the cost of carrying it out.

  • Please rate the clarity and organization of this paper

    Very 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 mention open access to source code or data but provides a clear and detailed description of the algorithm to ensure reproducibility.

  • Do you have any additional comments regarding the paper’s reproducibility?

    N/A

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html

    As the physical parameters can also change over time and environment, is this methodology and the associated training applied to each individual acquisition? In this case, the time cost is critical for its feasibility.

  • 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

    Accept — should be accepted, independent of rebuttal (5)

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    The presented work is show a novel a pysics-guided material decomposition model for PCCT, which reached better results compared to the state of the art works. In addition the work is well-written and interesting for this communicaty

  • Reviewer confidence

    Confident but not absolutely certain (3)

  • [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

    N/A

  • [Post rebuttal] Please justify your decision

    N/A



Review #4

  • Please describe the contribution of the paper

    In this paper, the authors present a deep learning model for modeling the detector and ASIC of photon-counting CT (PCCT). The proposed network is trained using the simulation data from a Monte Carlo model and then calibrated to real PCCT scanner. The results indicate that the proposed method can effectively reduce the image artifact and lead to better CT reconstruction.

  • Please list the main strengths of the paper; you should write about 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 proposed method integrates extensive physics knowledge and thus the training of the network is guided by the physics parameters.
    2. Although the parameters can be obtained by Monte Carlo model, the proposed network is hopefully less time-consuming and suitable for image reconstruction.
  • Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.

    Not applied.

  • Please rate the clarity and organization of this paper

    Excellent

  • 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.

  • Do you have any additional comments regarding the paper’s reproducibility?

    The data used for training is from Monte Carlo simulation and thus reproduction is perhaps impossible.

  • Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html

    Not applied

  • 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

    Accept — should be accepted, independent of rebuttal (5)

  • Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

    The paper is quite well written and the presented results are promising. The physics part is difficult for some readers, but the authors put efforts to simplify the modeling, making it easier to follow

  • Reviewer confidence

    Somewhat confident (2)

  • [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed

    N/A

  • [Post rebuttal] Please justify your decision

    N/A



Author Feedback

We greatly appreciate reviewers for their detailed feedback and insightful suggestions. Below, we address each of the raised concerns:

Use of Maximum Log-Poisson Loss and Network Output (R1, R3): Under the X-ray imaging flux we studied, the system output counts follow a Poisson distribution, as detection events occur independently at a constant mean rate. This is a standard assumption in the field [1]. In our study, the Detector Net predicts the full spectrum, while the ASIC Net calculates expected bin counts. We will add more annotations in Figure 1 for clarity.

Concerns about Improvements Shown in Table 1 (R1): Our method focuses on accurately modeling the PCCT system and reducing bias and structural artifacts. Denoising is not our main focus. As Table 1 shows, our method can achieve a significant reduction in bias, maintaining within 1 HU.

Monte Carlo Model, Open Source Availability & Reproducibility(R3, R4, R5): Our Monte Carlo model integrates elements from several established methodologies. We generate the initial spectrum using [2], incorporate bowtie-filter physics from clinical settings, and utilize energy deposition parameters from [3]. Charge transportation and signal generation modeling follows [4-5], and active-reset ASIC architecture is based on [6]. We are open to providing additional guidance to researchers attempting to replicate our work. To support community research, our simulation code is available upon request via email.

Optimization and Parameter Estimation Methodology (R3, R4): For estimating parameters such as dead time in Equation (7), we use calibration data with known material depths to find the optimal parameters that minimize the Poisson loss. Similarly, for estimating material depths in Equation (8), we identify the optimal material depths that minimize the Poisson loss. We utilize the Adam optimizer in PyTorch for these tasks. The full spectrum of PCCT is derived using the physics parameters estimated (such as bowtie filter depth) through this process.

Calibration Frequency and Time Costs (R4): In typical clinical environments, PCCT systems undergo calibration every 2-3 weeks. Our model supports simultaneous calibration of all detector pixels, with each model being compact and independent (approximately 1Mb). Material decomposition follows a similar rapid process. Ideally, calibration process should be finished within 1 hour and material decomposition should be finished within minutes. However, current practical implementations of our model require approximately one hour for both calibration and material decomposition due to the preliminary nature of our parallel processing tool, which needs further optimized.

[1] Wang et al. Sufficient statistics as a generalization of binning in spectral x-ray imaging. IEEE TMI 2010 [2] Punnoose et al. spektr 3.0—A computational tool for x‐ray spectrum modeling and analysis. Medical physics 2016 [3] Jan et al. GATE V6: a major enhancement of the GATE simulation platform enabling modelling of CT and radiotherapy. Physics in Medicine & Biology 2011 [4] Lai et al. Modeling photon counting detector anode street impact on detector energy response. IEEE TRPMS 2020. [5] Taguchi et al. Spectral, photon counting computed tomography: technology and applications. CRC Press, 2020. [6] Knoll et al. Radiation detection and measurement. John Wiley & Sons, 2010.



Meta-Review

Meta-review not available, early accepted paper.



back to top