List of Papers Browse by Subject Areas Author List
Abstract
The paraspinal muscles are crucial for spinal stability, which can be quantitatively analyzed through image segmentation. However, unclear muscle boundaries, severe deformations, and limited training data impose great challenges for existing automatic segmentation methods. This study proposes an automated probabilistic inference framework to reconstruct 3D muscle shapes from thick-slice MRI robustly. Leveraging Fourier basis functions and Gaussian processes, we construct anatomically interpretable shape models. Multi-level contextual observations such as global poses of muscle centroids and local edges are then integrated into posterior estimation to enhance shape model initialization and optimization. The proposed framework is characterized by its intuitive representations and smooth generation capabilities, demonstrating higher accuracy in validation on both public and clinical datasets compared to state-of-the-art methods. The outcomes can aid clinicians and researchers in understanding muscle changes in various conditions, potentially enhancing diagnoses and treatments.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/1932_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{Wan_Automated_MICCAI2024,
author = { Wang, Jinge and Chen, Guilin and Wang, Xuefeng and Wu, Nan and Zhang, Terry Jianguo},
title = { { Automated Robust Muscle Segmentation in Multi-level Contexts using a Probabilistic Inference Framework } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
year = {2024},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15009},
month = {October},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper proposes the probabilistic inference framework for reconstruction of 3D para-spinal muscle shapes using multi-level contextual information. Pose correlation, probabilistic shape model and edge confidence have been utilized to represent the anatomical features. Proposed probabilistic shape model based on Fourier basis functions and Gaussian processes correlates the local shape details and vertebral ID’s of the muscle respectively. In case of severe fat infiltration, the topological relationship between the muscles has been represented by computing approximate center positions in global pose. This work demonstrates the comprehensive segmentation and reconstruction method for 3D para-spinal muscles using interpretable shape models.
- 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.
a) 3D muscle reconstruction method is proposed from multi-level contextual information using global pose,probabilistic shape models and edge confidence.
b) The proposed approach requires minimal parameters to represent the muscles without the need of landmark alignment through registration.
c) The probabilistic inference framework has been proposed to find out the shape model parameters through Maximum A Posteriori (MAP) estimation using edge observations.
d) All the mathematical expressions consisting the final kernel function,curve confidence and Maximum A Posteriori (MAP) distribution are well explained.
- 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.
a) Please elaborate the role of dynamic noise kernel in probabilistic shape model.
b) Please illustrate dataset details: data collection ( public dataset link, clinical dataset hospital name etc), image acquisition and data processing details.
c) Please provide the intermediate results for the computed multi-level contextual information such as pose detection, segmentation and edge detection.
- 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 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?
Please provide the details for the ablation experiments for the factors edge information and dynamic noisy observation.
- 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
The paper proposed a novel technique for the 3D reconstruction of para-spinal muscles using an interpretable shape model. Multi-level contextual information observed with pose, edge and shape models is useful for understanding the anatomical features. However, the methodological illustration can be provided about the two YOLOv8-s models used for pose and segmentation observations. The optimization technique for the network and hyper-parameters of the network can be clearly mentioned.
- 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 proposed novel technique for muscle reconstruction with interpretable shape modeling using pose, edge confidence can be useful to find out the topological relationship in case of severe fat infiltration. The suggested comprehensive reconstruction and segmentation method can be useful to generate anatomically correct 3D para-spinal muscles.
- 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 #2
- Please describe the contribution of the paper
This paper proposes a novel probabilistic inference framework for muscle segmentation. The authors novelly represent the 3D muscle as multiple radially contours stacked axially and utilize Gaussian process to model them. The SOTA results from both public and clinical datasets demonstrate the effectiveness.
- 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 is a novel method for muscle segmentation. The way to represent and model muscles is interesting.
- The paper is well-written and well-organized.
- 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.
- Due to the complexity of the method, the provided details are insufficient for reproducibility, such as Eq. 4. It would be beneficial if the authors could include additional details in the supplementary material or release the codes to facilitate better replication of the approach.
- The term “multi-level contexts” requires further elaboration. Additional explanations or descriptions would be helpful in clarifying the concept.
- 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 provide sufficient information for reproducibility.
- Do you have any additional comments regarding the paper’s reproducibility?
The details of the method part is insufficient for reproducibility. Releasing the source codes is encouraged.
- 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
Please address my concerns in the weakness part.
- 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 well-written and well-organized. The proposed method appears to be reasonable and of novelty.
- 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 #3
- Please describe the contribution of the paper
The paper proposes a method to fully utilize anatomical context at different levels like pose, shape and edges for the task of paraspinal muscle segmentation
- 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.
Paper is very well written with ample visuals to understand the proposed method. It compares against strong baselines like the nnUNet.
- 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.
No obvious weaknesses
- 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?
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
Please consider making the code public, it’ll help the community evaluate the approach on tasks of similar grain.
- 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 proposed method exhibits great potential, as evidenced by the results.
- 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 thank all three reviewers for their constructive comments and appreciations of our strengths. We carefully consider each point raised by the reviewers and make the following clarifications: [R1] The role of dynamic noise kernel: The dynamic noise kernel is essentially a noise kernel that reflects the inherent uncertainty of the observations. In our work, influenced by factors such as dataset size, image complexity, and inter-annotator consistency, segmentation observations at different slices are considered to have varying degrees of confidence, with manual observation confidence set as 1. This dynamic noise results in a larger search space for edge optimization when automatic segmentation is less reliable, thereby ensuring robustness. [R1] Details about the datasets: The public dataset comes from the following paper which contains the link:Burian et al.: Lumbar muscle and vertebral bodies segmentation of chemical shift encoding-based water-fat MRI: the reference database MyoSegmenTUM spine. (BMC Musculoskelet. Disord. 2019).
The clinical dataset from Peking Union Medical College Hospital (PUMCH) comprises a total of 46 cases of spinal deformities (19 males and 27 females, age: 19.6±9.9 years). These cases were meticulously annotated using 3D Slicer. In each case, lumbar vertebrae were annotated in the T1-SAG sequence, while the erector spinae, psoas, and multifidus muscles were annotated on all slices in the T2-AXI sequence. [R1] Intermediate results: We will supplement the intermediate results of pose, edge, and segmentation in Figure 3 of the final version of our paper. [R1] Details about ablation experiments: Ours-Dyn.Noise-Edge opt.: This group uses the results of pose detection and automatic segmentation as observations, with the confidence always set to 1, initializes the probabilistic shape model, and takes this as the final segmentation result. Ours-Dyn.Noise: This group, building on the previous one, trains a muscle edge detector and optimizes the geometric parameters of the probabilistic shape model at each slice during inference. Ours: This group, building on the previous one, assumes that the confidence levels of observations at different slices are different. [R1] The methodological illustration of pose detection and segmentation model: These models are based on the Ultralytics’ Yolov8-s-pose and -seg architectures. The image size is set to 512x512, with the optimizer being Adam, an initial learning rate of 0.01, momentum of 0.937, for 400 epochs, a batch size of 16, without the use of data augmentation techniques, and all other parameters are set to default. [R4&R5] Code release: Our code is divided into two parts: pose and edge detection, along with preliminary segmentation, are handled in Python and saved as .mat files. These files are then read by Matlab, where the probabilistic shape model is built, trained, and optimized. We intend to clean up and perfect the code and open-source it before submitting the expended version of our work to a journal. For replication purposes, the muscle representation’s clear structure should make it easy to replicate using available Fourier transform and GP libraries in different programming languages. [R4] The clarification for the multi-level context: “Multi-level context” emphasizes the importance of integrating information from various sources and levels of detail. This phrase has appeared in works such as Li et al.’s “A Multi-Level Contextual Model For Person Recognition in Photo Albums” (CVPR, 2016) and Liu et al.’s “Multi-level context-adaptive correlation tracking” (PR, 2019). In this work, we exploit three levels of context for robust muscle segmentation: a) spine-level, which consists of the muscles’ relative position and anatomical height features; b) edge-level, including muscle boundaries and the confidence of elective muscle shape; c) deep-level, which presents deep features of muscle images that are not easily established through intuitive cognition.
Meta-Review
Meta-review not available, early accepted paper.