List of Papers Browse by Subject Areas Author List
Abstract
Recent studies have proposed quantitative ultrasound (QUS) to extract the acoustic properties of tissues from pulse-echo data obtained through multiple transmissions. In this paper, we introduce a learning-based approach to identify thyroid nodule malignancy by extracting acoustic attenuation and speed of sound from ultrasound imaging. The proposed method employs a neural model that integrates a convolutional neural network (CNN) for detailed local pulse-echo pattern analysis with a Transformer architecture, enhancing the model’s ability to capture complex correlations among multiple beam receptions. B-mode images are employed as both an input and label to guarantee robust performance regardless of the complex structures present in the human neck, such as the thyroid, blood vessels, and trachea. In order to train the proposed deep neural model, a simulation phantom mimicking the structure of human muscle, fat layers, and the shape of the thyroid gland has been designed. The effectiveness of the proposed method is evaluated through numerical simulations and clinical tests.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/0306_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{Kim_Quantitative_MICCAI2024,
author = { Kim, Young-Min and Kim, Myeong-Gee and Oh, Seok-Hwan and Jung, Guil and Lee, Hyeon-Jik and Kim, Sang-Yun and Kwon, Hyuk-Sool and Choi, Sang-Il and Bae, Hyeon-Min},
title = { { Quantitative Assessment of Thyroid Nodules through Ultrasound Imaging Analysis } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
year = {2024},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15004},
month = {October},
page = {pending}
}
Reviews
Review #1
- Please describe the contribution of the paper
The paper presents a deep learning method to estimate quantitative parameters of attenuation and speed of sound.
- 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 not only strong in synthetic data generation (considering different inclusions and having simulation data similar to thyroid data), but also benefits from updated network blocks.
- The ablation study demonstrates the improvements of added blocks.
- The invivo results can differentiate benign and malignant thyroid tumors.
-
- 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.
-other parts of the neural network architecture could be investigated in ablation study (example: SPADE)
- 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?
nan
- 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 original SPADE, the mask is used to find the scaling of features, here the authors used RF for scaling of Bmode image. RF data has the main feature to estimate attenuation and backscattering. Why in equation (1), the RF data features were used for scaling of Bmode features? I think it should be opposite (Bmode determines the scaling of RF features)
- 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 has sevral strength with a few weaknesses.
- 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 #2
- Please describe the contribution of the paper
The authors propose a novel encoder-decoder architecture, called QIT-net, for quantitative ultrasound (QUS) imaging, specifically for thyroid nodules. The QIT-net incorporates a CNN-Transformer hybrid encoder for ultrasound (US) radio frequency (RF) signals and another CNN encoder for B-mode images. The extracted features from RF signals and B-mode images are fused utilizing the SPADE layer and decoded to generate maps of acoustic attenuation and sound speed. Additionally, B-mode image reconstruction is utilized as an auxiliary task to enhance robust performance. Alongside the novel architecture, a realistic US simulation phantom is designed by employing CT images for creating a realistic simulation-based training dataset. Overall, the authors demonstrate that the proposed method can improve the accuracy of neural network based QUS imaging through comprehensive evaluations.
- 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.
- Novelty: The QIT-net adopts a SPADE layer for fusing features extracted from RF signals and B-mode images. With this approach, it is expected that semantic image synthesis performance is improved, as SPADE was originally proposed for this purpose. The auxiliary task of B-mode image reconstruction is another interesting idea for enhancing robust performance, and the authors demonstrated its beneficial effect. Additionally, the realistic phantom generation, which can capture miscellaneous anatomical details, was advantageous for in vivo scenarios where anatomical objects have complex shapes.
- Well-organized experiment materials: The proposed method was validated by training on simulation dataset and evaluating on clinical dataset. Based on the paper, the simulation dataset was constructed to simulate real in vivo condition as much as possible.
- Expandability: Although the QIT-net was proposed specifically for quantitative ultrasound (QUS) imaging of thyroid nodules, the method has potential for application in other diseases as well as other ultrasound-related tasks, such as segmentation, image enhancement, beamforming, and others, considering its capability to utilize large pre-beamformed radio frequency (RF) signal data and reconstruct B-mode images.
- 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.
Overall, the proposed method is well-validated. However, some additional details would be beneficial for further understanding. For instance, while the reference [5] is provided for determining the speed of sound and attenuation coefficient in the ultrasound simulation, it is unclear how the scatter diameter and scatter concentration were determined. Additionally, a potential limitation of this work is the generalizability of the method to other ultrasound probes beyond the one utilized for training and testing. Since the QIT-net utilizes pre-beamformed radio frequency (RF) signal data, its performance might be severely degraded when applied to probes with different characteristics, such as element count, kerf size, center frequency, and shape (linear or convex), as these factors significantly influence the RF signal characteristics.
- 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?
If it is available, release of the source code and dataset would be appreciate for the research community. Especially, the dataset or the phantoms of the dataset wolud be beneficial regarding the authors’ effort for making them.
- 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
I believe this work is interesting and acceptable in a conference format. However, for future journal publication, several concerns should be discussed: Limited Related Work and Comparisons: The related work and compared methods are limited in this study. It would be more informative if the authors introduce more deep learning-based methods for estimating speed of sound (SoS) and attenuation coefficients as related works, and compare the proposed method with these methods. Domain Gap due to Probe, Machine, or Imaging Settings: The potential domain gap caused by using different ultrasound probes, machines, or imaging settings should be acknowledged and discussed. This is a crucial limitation that could affect the generalizability of the method. The recently published transfer function-based method by Soylu and Oelze [1] would be a promising candidate for addressing this limitation. The authors should consider discussing this approach and exploring its potential for improving the generalization capability of the proposed QIT-net across different domains. Overall, while the work is interesting and suitable for a conference format, addressing these concerns related to comprehensive comparisons with other deep learning methods and the domain gap due to varying imaging conditions would significantly strengthen the work for future journal publication. Incorporating discussions and potential solutions, such as the transfer function-based method, would make the work more robust and enhance its impact. [1] Soylu, Ufuk, and Michael L. Oelze. “Machine-to-Machine Transfer Function in Deep Learning-Based Quantitative Ultrasound.” IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control (2024).
- 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 authors proposed CNN-transformer hybrid architecture and B-mode image reconstruction auxillary task for a quantitative ultrasound. The idea is novel enough and validated well.
- 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
This study proposes a novel neural network, QIT-net, for quantitative ultrasound (QUS). QIT-net leverages a combination of a convolutional neural network (CNN) for analyzing intricate local pulse-echo patterns and a Transformer architecture for capturing complex relationships between multiple beam receptions. Notably, ultrasound B-mode images serve as both input and label, enhancing the network’s accuracy. The researchers trained QIT-net on simulated phantoms mimicking soft tissue and thyroid characteristics. Additionally, they evaluated its performance on real in-vivo ultrasound data from benign and malignant tumors.
- 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 work addresses the need for accurate QUS methods by exploring transformer-based techniques and utilizing tissue-mimicking phantoms for training. Quantitative results demonstrate improvements achieved with QIT-net.
- 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 authors tested their method on in-vivo data, interpreting the imaging results is challenging due to the absence of ground truth information.
- 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.
- 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
While the authors tested their method on real in-vivo data, interpreting the imaging results is challenging due to the absence of ground truth information. Testing on phantoms with known sound speed (SOS) and attenuation (ATT) values, such as agar inclusions in water at controlled temperatures, would strengthen the validation process. The current quantitative metrics used in this study are 1-dimensional, failing to provide comprehensive information on the benefits of QIT-net. The authors could consider employing 2D- quantitative maps, like local SSIM maps, for a more holistic evaluation.
- 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?
Accurately imaging acoustic properties within a medium has long been a hurdle. This study tackles this challenge by proposing a novel neural network solution. Trained on realistic phantom data, the network leverages transformer architecture and utilizes ultrasound B-mode images as both input and output labels. This innovative approach demonstrates the potential for achieving accurate results when applied to real-world in-vivo ultrasound data.
- 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
Author Feedback
N/A
Meta-Review
Meta-review not available, early accepted paper.