List of Papers By topics Author List
| Paper Info | Reviews | Meta-review | Author Feedback | Post-Rebuttal Meta-reviews |
Authors
Weihang Liao, Yinqiang Zheng, Hiroki Kajita, Kazuo Kishi, Imari Sato
Abstract
Photoacoustic imaging (PAI) is a newly emerging bimodal imaging technology based on the photoacoustic effect; specifically, it uses sound waves caused by light absorption in a material to obtain 3D structure data noninvasively. PAI has attracted attention as a promising measurement technology for comprehensive clinical application and medical diagnosis. Because it requires exhaustively scanning an entire object and recording ultrasonic waves from various locations, it encounters two problems: a long imaging time and a huge data size. To reduce the imaging time, a common solution is to apply compressive sensing (CS) theory. CS can effectively accelerate the imaging process by reducing the number of measurements, but the data size is still large, and efficient compression of such incomplete data remains a problem. In this paper, we present the first attempt at direct compression of incomplete 3D PA observations, which simultaneously reduces the data acquisition time and alleviates the data storage issue. Specifically, we first use a graph model to represent the incomplete observations. Then, we propose three coding modes and a reliability-aware rate-distortion optimization (RDO) to adaptively compress the data into sparse coefficients. Finally, we obtain a coded bit stream through entropy coding. We demonstrate the effectiveness of our proposed framework through both objective evaluation and subjective visual checking of real medical PA data captured from patients.
Link to paper
DOI: https://link.springer.com/chapter/10.1007/978-3-031-16446-0_53
SharedIt: https://rdcu.be/cVRTN
Link to the code repository
N/A
Link to the dataset(s)
N/A
Reviews
Review #1
- Please describe the contribution of the paper
In this paper, a graph-based compression scheme is proposed for incomplete 3D photoacoustic data. Both objective and subjective evaluations are provided to demonstrate the compression quality.
- 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) As claimed by the authors, it is the first work to apply graph signal processing in incomplete PA compression. (2) The proposed method allow acceleration of data acquisition and small data size. (3) Medical doctors are invited to assess the proposed method.
- 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) The benefit and significance of compressing PA data is not elaborated in the paper, and I think it is a major issue to support the value of this paper. (2) Experimental results need more explanation. For example, I noticed the authors claimed a larger compression ratio K brings more advantages of the proposed method. But the results shown in Supplementary Material seem not very convincing. (3) The clarity and accuracy of the writing language needs improvement.
- Please rate the clarity and organization of this paper
Satisfactory
- Please comment on the reproducibility of the paper. Note, that authors have filled out a reproducibility checklist upon submission. Please be aware that authors are not required to meet all criteria on the checklist - for instance, providing code and data is a plus, but not a requirement for acceptance
Although related codes are not publicized, I believe this method is not difficult to reproduce with introduced technical details.
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://conferences.miccai.org/2022/en/REVIEWER-GUIDELINES.html
(1) The significance of the proposed method should be stressed by providing more explanation on PA data compression. Specifically, what is the data size you used in the article? If the (W,H,D) in Section 2.1 are not large, so is such compression valuable? (2) I noticed a conventional CS method (BP algorithm) is used here. I wonder if there is other state-of-art CS can be used here? Because the sampling ratio is not very low here, meaning regular CS methods are sufficient to recover the signals confidently. (3) Words and presentations should be chose more carefully. For example, in Section 2.1, is the equation “sigma X_=P_inc” an accurate formulation? Besides, the authors claimed “As medical data tends to be sparse, …”, general medical data is sparse in what sense? (4) In Equation (3), only the largest $\lambda$ is used, what if we keep top N largest values? (5) In Figure 3, I observe the margin between “DCT-only” and “GBC-only” is narrower compared with the one beween “Prop-full” and “GBC-only”? If combined with the concern about the rationale of PA data compression. This point further determines the merit of this paper.
- Rate the paper on a scale of 1-8, 8 being the strongest (8-5: accept; 4-1: reject). Spreading the score helps create a distribution for decision-making
4
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
Based on the current presentation of the significance of this paper and the experimental results, I give my recommendation.
- Number of papers in your stack
5
- What is the ranking of this paper in your review stack?
4
- Reviewer confidence
Somewhat Confident
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
Not Answered
- [Post rebuttal] Please justify your decision
Not Answered
Review #2
- Please describe the contribution of the paper
The work proposes a tailored compression approach for photoacoustic data. The compression mode is locally adaptive, depending on present features and in the advanced mode uses graphs to encode features.
- 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 compression shows good performance compared to classic compression. Compression performance has been tested with clinical data and experts rated that there is no clinical value lost with the compressed data.
- 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.
Some technical details are unclear to me, such as how the image data is turned into the graph structure is not entirely clear. I believe it would not be straight forward to reproduce the results. Nevertheless, the authors have indicated that code will be released, so this would help.
The premise of compression of undersampled data in the reconstruction is not clear. The undersampling would appear in the PA measurements and the image domain is usually fully sampled. This needs more detail or motivation
I would have hope also for more computational details to be included: needed memory in absolute (MB) and not relative. Computation times etc.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Note, that authors have filled out a reproducibility checklist upon submission. Please be aware that authors are not required to meet all criteria on the checklist - for instance, providing code and data is a plus, but not a requirement for acceptance
Codes will be released. Otherwise reproducibility would be difficult.
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://conferences.miccai.org/2022/en/REVIEWER-GUIDELINES.html
My primary concern is how the undersampling and reconstruction is performed. Usually in PA and other tomographic applications, the undersampling to save measurement time is performed in the measurement space (here the PA signal, which is a time-series). Reconstructions are then performed in the image domain, there is no undersampling done here in the reconstructions. If I understand the paper correctly, the authors state that an undersampling is performed in the reconstructed PA image (section 2.1). This does not make sense to me. The undersampling needs to be performed in the measurement space. Additionally, the undersampling would not be random and follows a pattern that pixels on the 2D scanning domain are omitted, while for all measured points the full time series is available.
This needs more clarification and motivation. Maybe I did not understand it and the undersampling is just done to save memory of storing the image? But then the whole discussion on compressed sensing (which concerns measurement space) is misleading.
For the construction of the graph, it is not entirely clear how we go from the voxel grid to the graph. This possibly assumes the random omission of voxels? These would be then excluded in the graph? Maybe a figure could help to illustrate this. In equation (1), could you explain the role of \sigma?
I would ask for some computational details. How long does the encoding part in the compression take and also the decoding? Could you add some absolute numbers, MB of image before and after compression?
- Rate the paper on a scale of 1-8, 8 being the strongest (8-5: accept; 4-1: reject). Spreading the score helps create a distribution for decision-making
3
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
I believe the premise for the undersampling in image domain is either incorrect or not well explained.
- Number of papers in your stack
5
- What is the ranking of this paper in your review stack?
4
- Reviewer confidence
Confident but not absolutely certain
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
5
- [Post rebuttal] Please justify your decision
My previous score was based on misunderstanding the data acquisition. The clarification that the potential undersampling will lead to an equivalent reduction in scan time increases relevance considerably. Thus I am willing to change my score to 5+ (weak accept with tendency to 6 accept, but not quite due to current weaknesses in the technical details)
Review #3
- Please describe the contribution of the paper
The paper approaches the problem of incomplete 3D PA in a graph-based encoding scheme. A reliability-aware rate-distortion optimization (RDO) is proposed to enable adaptive compression PA observations based on different reliability levels.
- 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 topic is interesting and clinically significant.
- The paper is well organized and easy to follow.
- A series of experiments in real clinical data demonstrated the superiority of the proposed method over DCT.
- 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 proposed reliability-aware rate-distortion optimization (RDO) need try all three coding modes and then calculate the coding cost, which significantly increase the overhead of compression.
- Comparison is limited to Graph-based methods. While authors point out several disadvantages of popular image/video compression standards, there are not experiments to support it.
- The detail of decoding is missing in method part.
- Please rate the clarity and organization of this paper
Very Good
- Please comment on the reproducibility of the paper. Note, that authors have filled out a reproducibility checklist upon submission. Please be aware that authors are not required to meet all criteria on the checklist - for instance, providing code and data is a plus, but not a requirement for acceptance
Authors do not mention the availability of source code.
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review: https://conferences.miccai.org/2022/en/REVIEWER-GUIDELINES.html
-
Including more comparisons with popular image/video compression standards and other learning-based methods will make this approach more convincing. For example, AVC [27], HEVC [20], ref A, and ref. B. Ref A: Pengfei Guo, Dawei Li, and Xingde Li, “Deep OCT image compression with convolutional neural networks,” Biomed. Opt. Express 11, 3543-3554 (2020) Ref B: Zhihao Hu, Guo Lu, Dong Xu; “FVC: A New Framework Towards Deep Video Compression in Feature Space” Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2021, pp. 1502-1511
-
Authors may clarify the relation between Q and compression ratio (CR)? What’s the definition of the compression ratio? From Fig.3, under same Q, the proposed method always achieves a smaller compression ratio.
-
- Rate the paper on a scale of 1-8, 8 being the strongest (8-5: accept; 4-1: reject). Spreading the score helps create a distribution for decision-making
5
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The rating reflects innovation of the approach for 3D PA compression.
- Number of papers in your stack
3
- What is the ranking of this paper in your review stack?
1
- Reviewer confidence
Very confident
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
Not Answered
- [Post rebuttal] Please justify your decision
Not Answered
Primary Meta-Review
- Please provide your assessment of this work, taking into account all reviews. Summarize the key strengths and weaknesses of the paper and justify your recommendation. 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.
Reviewers appreciated that this work provides the first tailored compression approach for photoacoustic data, and that it achieves good compression rates. However, they were missing information about absolute file sizes and the computational effort in order to better judge the practical relevance of the proposed codec. Moreover, the relative roles of acquiring incomplete data and the proposed compression, and the technical details of graph construction did not become clear enough. Even though the current numerical scores suggest rejection, I believe it might be possible to address these concerns in a rebuttal.
- What is the ranking of this paper in your stack? Use a number between 1 (best paper in your stack) and n (worst paper in your stack of n papers). If this paper is among the bottom 30% of your stack, feel free to use NR (not ranked).
5
Author Feedback
We thank all the reviewers for the detailed feedback. We address the main concerns as follows. Data size and compression significance (R1, R2): The resolution of a full leg PA data is 2176x1440x256, stored in 16-bit depth, and the full data size is 1.5GB. In our experiment, we cropped out 400x400 local areas, thus W=H=400, D=256, and the data size of experiment input is 78MB. As we reported in section 3.3, Fig.4, we showed the reconstructed data (K=0.5, Q=100) to doctors. Under this case the CR is 5.4%, the compressed bit-stream is only about 4.2MB, while the doctors cannot distinguish the visual differences. We believe the compression significance should be well demonstrated through this experiment. Computational cost and benefit (R1, R2, R3): We would like to emphasize that, our approach can reduce scanning time and data storage requirements, although it requires additional computation cost. The PAI system often suffers from a long data acquisition time. According to the doctors at the hospital we work with, it takes about 10 minutes to capture one PA image of a patient leg. It is a burden on patients, especially elderly patients, to keep still in scanning posture. Our framework enables us to reduce measurement points and leads to a shorter scanning time, thereby relieving the burden on the patient. Indeed, our method introduces some computational overhead (for a 400x400x256 input, K=0.5, Q=100, we test on a laptop with a 2.8GHz CPU, it takes about 30~40 minutes to finish the compression), but this is done off-line after data acquisition. We believe a shorter data acquisition time is more beneficial, as it is closely related to patient experience. We will add this to the Introduction to better justify our motivation. Sampling scheme (R2): We note that we mainly consider the scanning-based PAI modality, in which the object is scanned by moving a laser source and ultrasonic transducer along the surface. At each scan position, one 1D vector is reconstructed from raw signals, and full 3D data can be obtained by combining 1D vectors at all positions. To accelerate data acquisition, it is desired to skip some scan positions, resulting in incomplete 3D data. Working with ADVANTEST corporation, we are implementing a prototype with random skip pattern in the Hadatomo Z PAI system. In this work, simulated incomplete data is examined, as described in section 3.1. Graph construction (R2): As the random skip pattern is recorded, we know the position of reconstructed and missing voxels. We obtain the graph by traversing each present voxel and connecting it with other present voxels in the 2-hop neighborhood. Sigma in Eq.1 is a penalty factor for edge weight, which is set to 1 by default in our experiments. We will present the technical details more clearly in the final version. Extra comparison (R3): We note that the default compression standards are designed for natural image content (8-bit, full grid). They cannot directly apply to incomplete PA data (16-bit, with missed voxels). We first tried HEVC software (HM-15.0+RExt-8.0). We set all missed pixels to 0 and tested YUV400, 16bit configuration, but the software got the errors. It is out of our scope to debug the HM software. We further tried the latest VVC standard (VTM-11.2+HDRLib). Due to time limitation, we only compressed one 2D slice (as intra frame) with QP 0, YUV400, 16bit; the CR is 16.61%, and PSNR is 41.6742. We suspect the poor performance is because, the intra prediction cannot handle random missed voxels very well, thus the residual energy will be very large. We also carefully studied the two references recommended by R3. But as PA is a relatively new technology, there is currently no large-scale dataset available, thus difficult to perform learning-based methods. We will cite these two references and add discussion in the final version. For other comments raised by R1 and R3, we appreciate them and will improve the manuscript accordingly.
Post-rebuttal Meta-Reviews
Meta-review # 1 (Primary)
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
The rebuttal managed to clarify a misunderstanding in one of the original reviews, as well as the significance of the achieved results, which was initially questioned by the reviewers. In my personal opinion, it’s a strength of this paper that it would add to the diversity at MICCAI, in terms of methods and modalities. It is my impression that the main remaining concern is the unclear presentation. In the hope that the authors will improve this in their camera-ready version, I would support acceptance.
- After you have reviewed the rebuttal, please provide your final rating based on all reviews and the authors’ rebuttal.
Accept
- What is the rank of this paper among all your rebuttal papers? Use a number between 1/n (best paper in your stack) and n/n (worst paper in your stack of n papers). If this paper is among the bottom 30% of your stack, feel free to use NR (not ranked).
9
Meta-review #2
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
The work proposes a tailored compression approach for photoacoustic data. The compression mode is locally adaptive, depending on present features, and the advanced mode uses graphs to encode features. The topic is interesting and clinically significant. A series of experiments in real clinical data demonstrated the superiority of the proposed method over DCT. Moreover, the paper is well organized and easy to follow. In Rebuttal, the author emphatically explains the premise for the under-sampling in the image domain, making the reviewer clearer about the method. That is, it answers the reviewer’s questions well. Hence, I recommend accepting this submission.
- After you have reviewed the rebuttal, please provide your final rating based on all reviews and the authors’ rebuttal.
Accept
- What is the rank of this paper among all your rebuttal papers? Use a number between 1/n (best paper in your stack) and n/n (worst paper in your stack of n papers). If this paper is among the bottom 30% of your stack, feel free to use NR (not ranked).
10
Meta-review #3
- Please provide your assessment of the paper taking all information into account, including rebuttal. Highlight the key strengths and weaknesses of the paper, clarify how you reconciled contrasting review comments and scores, indicate if concerns were successfully addressed in the rebuttal, and provide a clear justification of your decision. If you disagree with some of the (meta)reviewer statements, you can indicate so in your meta-review. Please make sure that the authors, program chairs, and the public can understand the reason for your decision.
The paper adresses an original problem: compression of 3D photoacoustic data with alternative SOTA tools to deep learning: graph signal processing. Despite the clarity issues that should be addressed in the final version of this paper, I support acceptance given the soundness of the approach, the improved results, the clarified clinical relevance, and the opportunity to broaden the scope of methods discussed in the MICCAI conference.
- After you have reviewed the rebuttal, please provide your final rating based on all reviews and the authors’ rebuttal.
Accept
- What is the rank of this paper among all your rebuttal papers? Use a number between 1/n (best paper in your stack) and n/n (worst paper in your stack of n papers). If this paper is among the bottom 30% of your stack, feel free to use NR (not ranked).
4