Comparative Study of Ultrasound Tissue Motion Tracking Techniques for Effective Breast Ultrasound Elastography

Round 1

Reviewer 1 Report

In this manuscript, the authors presents a comparison of four different theoretically sound displacement estimators (AM2D, GLUE,  OVERWIND and SOUL) for their ability in tissue Young’s modulus reconstruction. The effectiveness of each method was assessed. The diagnosis of the breast cancer is of great significance and the ultrasound detection has play a great role in clinical. The proposed AM2D combined with STREAL seems has the effects to some extent as the images show. The experiments of the comparisons are well done. Well, the manuscript still has some issues need to be revised and addressed before accepted, as I listed below.

1.    The red and blue lines with stars in the Figure 2 do not correspond one-to-one with the results whether STREAL was applied or not. Could you please provide more context or specific information about the data you are referring to?

2.    In Figure 3, the AM2D estimator provides a high quality reconstruction of the Young’s Modulus. While, besides the operation time, the reasons of other techniques about the inaccuracy need to be analyzed.

3.    In Table 1, there’s no obvious trends in AM2D and GLUE with or without the STREAL. The CNR and SNR of AM2D both decline. Is there evidence that it is better of the STREAL 's application is in AM2D?

4.    In Table 2 and 3, the data collected from mimicking phantom and clinical scenes show ideally trends of the effect of STREAL. Could you explain the detail difference about these data with those collected in-silico ultrasound?

5.    For the Figures 7 to 15, it’s indeed that the AM2D provided good results, but looks not for every patient the STREAL performed better when combined with AM2D. Could you please provide the details computational results like the Table 1 and 2 for these images?

Most of the languages in the manuscript is fine for scientific. 

Author Response

First, the authors would like to express their gratitude to the reviewer for his/her very valuable comments, critique, and constructive feedback. The comments s/he provided are indicative of his/her in-depth and thorough understanding of the paper’s content. We believe that addressing the given comments and the revisions we made to the manuscript has improved our original manuscript substantially. Please see below point-by-point responses to the comments.

Reviewer 1

In this manuscript, the authors present a comparison of four different theoretically sound displacement estimators (AM2D, GLUE, OVERWIND and SOUL) for their ability in tissue Young’s modulus reconstruction. The effectiveness of each method was assessed. The diagnosis of the breast cancer is of great significance and the ultrasound detection has play a great role in clinical. The proposed AM2D combined with STREAL seems has the effects to some extent as the images show. The experiments of the comparisons are well done. Well, the manuscript still has some issues need to be revised and addressed before accepted, as I listed below.

1. The red and blue lines with stars in the Figure 2 do not correspond one-to-one with the results whether STREAL was applied or not. Could you please provide more context or specific information about the data you are referring to?

Accepted: We thank you for bringing this to our attention, what we were referring to was indeed ambiguous. In the revise manuscript, we have clarified that the comparisons were done between displacement estimators with or without STREAL (i.e GLUE vs AM2D both with STREAL) rather than between the STREAL/NoSTREAL comparison. Please see the highlighted text labelled R1C1 in the revised manuscript.

2. In Figure 3, the AM2D estimator provides a high-quality reconstruction of the Young’s Modulus. While, besides the operation time, the reasons of other techniques about the inaccuracy need to be analyzed.

Accepted: We agree that there needs to be a more in-depth discussion on the sources of inaccuracies in OVERWIND and SOUL where the merits of each displacement estimator are given. This is now added to the manuscript. Please see the highlighted text labelled R1C2 in the revised manuscript

3. In Table 1, there’s no obvious trends in AM2D and GLUE with or without the STREAL. The CNR and SNR of AM2D both decline. Is there evidence that it is better of the STREAL 's application is in AM2D?

Accepted: Thank you for bringing this to our attention. Without more context, it does seem like the application of STREAL to AM2D is not very well motivated. The answer to this is that the differences in CNR and SNR in this table are relatively small, and the lack of large differences mostly comes down to the fact that the tissue mimicking phantom yields high-quality axial displacement estimates sufficient for high quality reconstruction even without STREAL. While it appears that this leaves only very little room for improvement in tissue displacement estimation in the axial direction, STREAL offers massive improvement in the lateral direction. Highlighting AM2D + STREAL stems from that the in-silico phantom where AM2D + STREAL was able to produce a comparable at worst and arguably better image at best while the clinical examples where AM2D + STREAL increases the prominence of the inclusion. This has been added in the Discussion section of the revised manuscript; please see the highlighted text labelled R1C3.

4. In Table 2 and 3, the data collected from mimicking phantom and clinical scenes show ideally trends of the effect of STREAL. Could you explain the detail difference about these data with those collected in-silico ultrasound?

Accepted: The difference between the data mentioned above stems from the fact that the in-silico phantoms are simulated with computational simulation software and represent a “perfect” scenario with no signal decorrelation. The data pertaining to the tissue mimicking phantom, however, was acquired with a real ultrasound probe and a controlled motion device. Moreover, the clinical data was acquired freehand from patients. The pronounced effect of STREAL in the tissue mimicking and clinical data is mostly because under the ideal conditions of in-silico phantoms, the high-quality axial displacements are largely sufficient for a high-quality reconstruction, hence STREAL’s performance becomes less obvious. In the clinical cases however, the inclusion of lateral displacement information in the reconstruction increases the quality of reconstruction due to increased information incorporated in the data inversion. In addition to the highlighted text labelled [R1C3], this has been now further clarified through adding the highlighted text labelled [R1C4] in the revised manuscript.

5. For the Figures 7 to 15, it’s indeed that the AM2D provided good results, but looks not for every patient the STREAL performed better when combined with AM2D. Could you please provide the details computational results like the Table 1 and 2 for these images?

Accepted: We thank you for your suggestion and have decided to include similar tables for the patients data that we included for the tissue mimicking phantom. However, we also cautioned readers that the interpretation of such metrics on the clinical data is difficult as the underlying heterogeneity and stiffness distribution is unknown. Please see Tables 6-8 and the highlighted text labelled [R1C5a&b] added in the revised manuscript.

Reviewer 2 Report

The real-time implementaion is desirable as future scope. Sensitivity and specificity and kappa static can be shown as a separate table

Minor English correction

Author Response

First, the authors would like to express their gratitude to the reviewer for his/her valuable comment, critique, and constructive feedback. We believe that addressing the given comment and the revisions we made to the manuscript has improved our original manuscript substantially. Please see below our response to the comment.

The real-time implementaion is desirable as future scope. Sensitivity and specificity and kappa static can be shown as a separate table.

Accepted: As indicated in the manuscript, typical runtime of AM2D is 0.1 s, which qualifies it for real-time method of tissue motion tracking. As for adding sensitivity and specificity kappa statistics, we thank the reviewer for suggesting a further analysis to strengthen our position. In response, we have implemented the measures as suggested. Please see the new Table 1 and highlighted text labelled [R2C1a&b] in the revised manuscript provided to explain the relevant methods and comment on the results.

Reviewer 3 Report

This paper conducts a comparative analysis of four theoretically robust displacement estimators, aiming to assess their suitability for real-time ultrasound elastography systems, with a particular focus on the quality-to-runtime ratio. The four methods under consideration, namely, AM2D, GLUE, OVERWIND, and SOUL, are evaluated both individually and in conjunction with the recently developed STREAL strain field enhancement technique, which is grounded in tissue mechanics-based regularization principles. The overarching goal is to determine which of these estimators offer optimal performance in the context of breast cancer diagnosis. While the paper effectively introduces the primary theme and objectives, there are several areas where further refinement and expansion are warranted:

(1) Section 1 provides an overview of AM2D, GLUE, OVERWIND, and SOUL methods; however, it would be beneficial to elaborate on why these particular displacement estimators were selected and what unique contributions this study brings to the field.

(2) Consider reorganizing the paper by separating Section 1 into a dedicated "Related Work" section (e.g., Section 2) to provide clearer motivation for the study.

(3) Due to the extensive use of abbreviations, consider including a comprehensive table in Section 1 to manage and explain these abbreviations for improved reader comprehension.

(4) In Section 3, elucidate the rationale behind the experimental design, including details on the data source and the justification for choosing these specific methods and techniques, which will enhance the clarity of the experimental setup.

(5) In the concluding section, it would be valuable to include a discussion of any limitations inherent in the study and to suggest potential future research directions within the domain of breast cancer diagnosis.

(6) Expand the introduction by providing a brief overview of various tools and approaches used in breast cancer diagnosis, such as uncertainty analysis and soft computing tools like D-S evidence theory and fuzzy theory, to offer readers a broader context for the study.

Author Response

First, the authors would like to express their gratitude to the reviewer for his/her very valuable comments, critique, and constructive feedback. The comments s/he provided are indicative of his/her in-depth and thorough understanding of the paper’s content. We believe that addressing the given comments and the revisions we made to the manuscript has improved our original manuscript substantially. Please see below point-by-point responses to the comments.

This paper conducts a comparative analysis of four theoretically robust displacement estimators, aiming to assess their suitability for real-time ultrasound elastography systems, with a particular focus on the quality-to-runtime ratio. The four methods under consideration, namely, AM2D, GLUE, OVERWIND, and SOUL, are evaluated both individually and in conjunction with the recently developed STREAL strain field enhancement technique, which is grounded in tissue mechanics-based regularization principles. The overarching goal is to determine which of these estimators offer optimal performance in the context of breast cancer diagnosis. While the paper effectively introduces the primary theme and objectives, there are several areas where further refinement and expansion are warranted.

• Section 1 provides an overview of AM2D, GLUE, OVERWIND, and SOUL methods; however, it would be beneficial to elaborate on why these displacement estimators were selected and what unique contributions this study brings to the field.

Accepted: Combined with the most recently developed STREAL method, the tissue motion tracking methods investigated in the manuscript were chosen for their being known to be ones of the most accurate tissue displacement estimators in ultrasound elastography literature. Moreover, their codes are publicly available while their theoretical foundations are substantially different. Some notable non-inclusions were mentioned in the discussion, namely NCCs and Deep Learning methods, as the former was of sufficiently inferior quality and runtime, hence it did not merit inclusion. The Deep Learning based methods would be unreliable to generalize and unfair to compare due to their GPU implementation. Major contributions of this study include comprehensive assessment of performance of state-of-the-art tissue displacement estimation methods and using the tissue Young’s modulus for this assessment as a true measure of stiffness compared to commonly used tissue strain which is not a true representative of stiffness. Please see the highlighted text labelled [R3C1] for an elaboration on the selection of the tissue motion estimators.

(2) Consider reorganizing the paper by separating Section 1 into a dedicated "Related Work" section (e.g., Section 2) to provide clearer motivation for the study.

Accepted: We thank the reviewer for helping us improve the structure of the paper. We have done as suggested and dedicated a separate section for related work. Please see the highlighted section labelled [R3C2a] in the revised manuscript. To further enhance the motivation of this work, we have appended the highlighted text labelled [R3C2b] to the Discussion section of the revised manuscript. The latter highlights potential future direction of this work, namely integrating the elastography technique identified to be highly effective into modern classification methodologies founded on Dempster-Shafer Theory or Fuzzy logic to improve the accuracy of breast cancer detection systems.

(3) Due to the extensive use of abbreviations, consider including a comprehensive table in Section 1 to manage and explain these abbreviations for improved reader comprehension.

Accepted: Thank you for your suggestion on how to improve the readability of the paper. Assuming that the suggestion is acceptable to the journal, we have added Table 1 to the revised manuscript which contains this information.

 (4) In Section 3, elucidate the rationale behind the experimental design, including details on the data source and the justification for choosing these specific methods and techniques, which will enhance the clarity of the experimental setup.

Accepted: Please see the highlighted text labelled [R3C4a-d] in the revised manuscript which provides the information suggested to be included in the above comment.

(5) In the concluding section, it would be valuable to include a discussion of any limitations inherent in the study and to suggest potential future research directions within the domain of breast cancer diagnosis.

Accepted: Added as suggested. Please see the highlighted text labelled [R3C5a&b] in the revised manuscript.

(6) Expand the introduction by providing a brief overview of various tools and approaches used in breast cancer diagnosis, such as uncertainty analysis and soft computing tools like D-S evidence theory and fuzzy theory, to offer readers a broader context for the study.

Accepted: Thank you for this excellent suggestion. For better flow, the suggested overview was included in the Discussion section of the revised manuscript before a potential future direction was suggested. Please see the highlighted text labelled [R3C6] in the revised manuscript.