Authors

Chi Ding, Qingchao Zhang, Ge Wang, Xiaojing Ye, Yunmei Chen

Abstract

We propose a novel Learned Alternating Minimization Algorithm (LAMA) for dual-domain sparse-view CT image reconstruction. LAMA is naturally induced by a variational model for CT reconstruction with learnable nonsmooth nonconvex regularizers, which are parameterized as composite functions of deep networks in both image and sinogram domains. To minimize the objective of the model, we incorporate the smoothing technique and residual learning architecture into the design of LAMA. We show that LAMA substantially reduces network complexity, improves memory efficiency and reconstruction accuracy, and is provably convergent for reliable reconstructions. Extensive numerical experiments demonstrate that LAMA outperforms existing methods by a wide margin on multiple benchmark CT data sets.

Link to paper

DOI: https://doi.org/10.1007/978-3-031-43999-5_17

SharedIt: https://rdcu.be/dnwwv

Link to the code repository

https://github.com/chrisdcs/LAMA-Learned-Alternating-Minimization-Algorithm

Link to the dataset(s)

N/A

Reviews

Review #2

Please describe the contribution of the paper

The authors propose a Learned Alternating Minimization Algorithm (LAMA) for CT image reconstruction using a variational model with learnable nonsmooth nonconvex regularizers. LAMA incorporates smoothing techniques and residual learning to reduce network complexity, improve memory efficiency, and enhance reconstruction accuracy, while being provably convergent for reliable reconstructions.
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.

-Theoretical proofs for convergence analysis and algorithms are presented. -Presents algorithm for nonconvex and nonsmooth minimization model. -Decent improvement in performance as compared to existing methods. -The method preserves structural details in reconstructed images while effectively suppressing noise and artifacts. The algorithm uses information from both domains to estimate missing information and enhance reconstruction quality by exploiting their complementary features.
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.

(10) in fig-1 refers to condition 10 from Algorithm-1 : could be added as fig description or text for clarity.
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

Supplementary document has some further proofs. However, No files to confirm the result/code reproducibility.
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/2023/en/REVIEWER-GUIDELINES.html

same as mentioned in the weakness. Overall it looks good work. Some insights into when to prescribe this proposed method vs the vanilla out of the shelf ML models should be really meaningful for the readers.
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

7
Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

Good theoretical work regarding nonconvex and nonsmooth minimization model. Very detailed proofs regarding the convergence and algorithm.
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

N/A
[Post rebuttal] Please justify your decision

N/A

Review #3

Please describe the contribution of the paper

This paper studied the CT image reconstruction problem using approach combining deep learning and alternating minimization where the deep learning is used to learn the regularization/prior knowledge of the ground truth while the alternating minimization provides updating rules for estimating.
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.

This paper looks quite interesting, and using deep learning to learn prior knowledge in reconstruction problem has been gaining great attention. This paper is well written, and empirical results indeed show some noticeable improvements over SOTA approaches over some datasets.
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 key steps are not well explained and reviewing of existing literature is less sufficient
Please rate the clarity and organization of this paper

Excellent
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

The authors promised to make implementations public
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/2023/en/REVIEWER-GUIDELINES.html
- In eq(2), when the authors used neural network to learn the regularization for the truth, they futher imposed $\ell_1,2$ regularization which does not sound intuitive to me. Can the authors provide more info about the motivations?
- In eq(6), the $f(x,z)$ is not defined. Should it be the first two terms in eq(1)?
- In eq(4)-(5), it does not look intuitive why the problem after the smoothing procedure is smooth. Also, the intuition of the smoothing procedure is also obvious to me.
- In [Yi et al., 2018], the authors also proposed to solve reconstruction problem using deep learning, alternating minimization, and the linear approximation technique. Can the authors clarify the difference between this work and [Yi et al., 2018]?
Yi J, Le AD, Wang T, Wu X, Xu W. Outlier detection using generative models with theoretical performance guarantees. arXiv preprint arXiv:1810.11335. 2018 Oct 26.
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

7
Please justify your recommendation. What were the major factors that led you to your overall score for this paper?

Novel idea and demonstrated noticeable improvement over SOTA approaches
Reviewer confidence

Very confident
[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

This paper uses a learnable variational model to solve the sparse angle CT reconstruction problem, and designs a novel adaptive scheme by modifying the alternating exist minimization methods to improve training efficiency.
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 mathematical derivation is relatively detailed. The optimization objective is decomposed into two stages and a learnable regularization term is incorporated, which enables effective minimization of the problem.
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 explanations of notations should be more detailed. The experiments of sparse view CT with noise have not be conducted.
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

The explanations of the notations are not detailed enough, which may hinder reproducibility.
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/2023/en/REVIEWER-GUIDELINES.html

(1) The explanations of the notations should be detailed enough. (2) The experiment of sparse view CT with noise should be conducted.
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?

This paper uses a learnable variational model to solve the sparse angle CT reconstruction problem. Mathematical derivation is relatively detailed but the explanations of the notations are not detailed enough, which may hinder the reproducibility and the understanding of the content.
Reviewer confidence

Very confident
[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

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.

The proposed method is new and interesting. The experimental results are good and support the proposed method.

Author Feedback

N/A

back to top

Learned Alternating Minimization Algorithm for Dual-Domain Sparse-View CT Reconstruction