A Rapid Head Organ Localization System Based on Clinically Realistic Images: A 3D Two Step Progressive Registration Method with CVH Anatomical Knowledge Mapping

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The current paper presents a technique for semi-automatic registration of TMI images to high definition predefined clinical data.

The technique may indeed be of practical interest to medical students or other groups of non-experts. The theoretical background (ICP) is very well known and has attracted a lot of attention with many variants through the years.

The main drawback of the paper is the lack of detailed description of the overall process of selecting points for the registration process in both stages.

Another point is the lack of objective measures for assessing the performance of the registration process in the results section.

Overall, the authors should focus on presenting the usefulness of the respective dedicated software rather than on the registration method (which is anything but new).

 

 

Comments on the Quality of English Language

English grammar and/or syntax is not correct in many cases. Also, some sentences seem to be incomplete. This may sometimes lead to incomplete understanding of concepts.

 

Author Response

Thank you for your valuable suggestion. We will reply to your questions one by one and mark the modified parts in red font in the revised draft, and remind you of the specific location in the reply

Comments 1:The main drawback of the paper is the lack of detailed description of the overall process of selecting points for the registration process in both stages.

Response 1:Thank you for your valuable comment. We understand the importance of point selection in describing the registration process in detail. In fact, our point selection in the first stage is based on the feature "eye nose triangle". The eye nose triangle is a relatively stable anatomical structure on the face that is not easily affected by changes in facial expression or other factors, and the eyes and nose are the most prominent features of the face that are easy to recognize and locate in most medical images. In the paper, we mentioned selecting 28 feature points with a discrete sampling frequency of 4 * 7 to balance computational efficiency and registration accuracy. By uniformly collecting 28 feature points on the respective surface contours, the method ensures that a sufficient number of points are used to capture the essential geometric features of the face without overwhelming the computational resources. Through experimental validation, this number of points is adequate to represent the facial features necessary for effective registration while maintaining a manageable computational load. It can well meet the requirements of the first registration's facial features for spatial orientation. It directly serves the first rigid registration (Facial features for spatial orientation) in Figure 4.

In Figure 5, what supports the spatial scale in the second step of similarity registration is the use of the facial circular contour, not the 28 feature points. Of course, these The 28 feature points evenly distributed across the face can serve as an important reference for selecting circular contours. Therefore, the 28 feature points are no longer directly relevant in the second step of similarity registration, which instead focuses on the similarity registration of the facial circular contour to determine the spatial scale.

If you are referring to how to select these feature points in the software operation process, We have uploaded a simple operation video for the process of selecting feature points in software operation, which may be more convenient to understand.

 

Comments 2:Another point is the lack of objective measures for assessing the performance of the registration process in the results section.

Response 2:Thank you for your valuable comment

We have added a quantitative analysis method based on ROI area coverage in section 3.4 of the results section to objectively evaluate the performance indicators of the registration process. Specifically, it includes:

(1) We invited a doctor with 12 years of work experience to annotate the typical anatomical structures (including optic nerve, pituitary gland, cerebellum, brainstem, and temporalis muscle) in 6 sets of data layer by layer, and had another doctor with 15 years of experience review to ensure the accuracy of ROI region division.

(2) We used a recall index to quantify the degree of overlap between the ROI regions automatically mapped by the evaluation system and the ROI regions annotated by doctors. The higher the recall value, the better the accuracy of positioning.

(3) We conducted detailed recall calculations and statistical analysis on data from 6 different modalities (incomplete-scan, thin-layer, or non-thin-layer CT, MRI, etc.) and different anatomical structures.

Overall, our proposed TSPR fast localization method ensures effective coverage of the ROI region and provides an objective quantitative evaluation of the registration process performance.

These quantitative analysis results provide objective basis for the performance evaluation of registration methods, and we hope to answer your questions.

 

Comments 3:Overall, the authors should focus on presenting the usefulness of the respective dedicated software rather than on the registration method (which is anything but new).

Response 3:Thank you for your valuable comment. We fully agree with your suggestion to pay more attention to the practicality of the specialized software we have developed. Therefore, we have added a new section (4.4 Clinical Application Prospects of the Software) specifically discussing the advantages of the usefulness of the dedicated software.

We hope that this modification can better demonstrate the practicality and clinical value of the software we have developed.

 

Finally, we sincerely thank the reviewers for their valuable feedback and constructive criticism. Your feedback not only helped us improve the quality of our paper, but also gave us deeper insights into our research. Your suggestions on increasing objective evaluation indicators and emphasizing software practicality are particularly valuable, as they greatly enhance the academic value and significance of our work. We greatly appreciate your valuable time and effort in reviewing our manuscript. Your professional insights have played an important guiding role in our research. We will continue to work hard to improve this research and hope to make more contributions to this field. Thank you again for your hard work and valuable suggestions.

Reviewer 2 Report

Comments and Suggestions for Authors

General Comments:

The paper deals with localization of Region of Interest in images and presents a tow-step method to localize ROI in tomographic medical images. The proposed approach showed good performance. The topic is an active area of research in image processing.

 

Specific Comments:

1.    On Page 6, Line 186, please clarify why registration of spatial orientation is performed only once.

2.    On Page 6, Lines 210-211, please justify the choice of 28 feature points (then, the discrete sampling frequency of 4*7). How would such a choice affect the image similarity registration in Figure 5?

 

3.    Please clarify how the proposed method can adapt to artefacts, changes, or incomplete information in ROI of medical images.

Author Response

Thank you for your valuable comments. We will reply to your questions one by one and mark the modified parts in red font in the revised draft, and remind you of the specific location in the reply

 

Comments 1:On Page 6, Line 186, please clarify why registration of spatial orientation is performed only once.

Response 1:Thank you for your valuable question. The reason why we only perform spatial orientation registration once is due to multiple factors:

（1）Meet the requirements:

The registration of spatial orientation is primarily to achieve rough spatial alignment between the TMIs and the CVH object. Once completed, subsequent fine registration (such as deformation registration) usually does not significantly change the overall spatial orientation.

（2）Calculation efficiency:

Spatial orientation registration typically involves global transformations such as rotation, translation, and scaling. These operations are relatively expensive in terms of computation. Performing only once can significantly reduce overall computation time, especially when dealing with large amounts of images or requiring real-time processing.

（3）Stability and avoid over registration:

Repeated executions of spatial orientation registration may lead to over registration, especially when processing medical images with local deformations. This may mask important anatomical differences, and increase the risk of local extremum problems.

（4）Hierarchical strategy:

This method adopts a hierarchical registration strategy. Firstly, perform rough global registration (spatial direction registration), followed by finer local registration. This strategy usually provides better overall results.

We have added a brief explanation of the reason for only performing spatial orientation registration once on page 6, lines 191-196.

 

Comments 2:On Page 6, Lines 210-211, please justify the choice of 28 feature points (then, the discrete sampling frequency of 4*7). How would such a choice affect the image similarity registration in Figure 5?

Response 2:Thank you for your question. The choice of 28 feature points, with a discrete sampling frequency of 4*7, is justified by the need to balance computational efficiency and registration accuracy. By uniformly collecting 28 feature points on the respective surface contours, the method ensures that a sufficient number of points are used to capture the essential geometric features of the face without overwhelming the computational resources. Through experimental validation, this number of points is adequate to represent the facial features necessary for effective registration while maintaining a manageable computational load. It can well meet the requirements of the first registration's facial features for spatial orientation. It directly serves the first rigid registration (Facial features for spatial orientation) in Figure 4.

In Figure 5, what supports the spatial scale in the second step of similarity registration is the use of the facial circular contour, not the 28 feature points. Of course, these The 28 feature points evenly distributed across the face can serve as an important reference for selecting circular contours. Therefore, the 28 feature points are no longer directly relevant in the second step of similarity registration, which instead focuses on the similarity registration of the facial circular contour to determine the spatial scale.

 

Comments 3:Please clarify how the proposed method can adapt to artefacts, changes, or incomplete information in ROI of medical images.

Response 3:Thank you for your question. The method proposed in this article has good adaptability to incomplete and non-thin layer quality real clinical data. It can adapt to the deformation and incomplete problem of ROI and the local extremum problem that is prone to occur in inter individual registration. For example, in Data 1 and 2, even if the optic nerve behind the orbital ball underwent greater deformation, or in Data 3, even if the invasion of the nodular shadow on the left side of the saddle area is more extensive, or in Data 4, even if the postoperative cerebellar changes underwent greater deformation, or in Data 5, even if teeth were missing, the method still achieved good registration in the area above the "eye nose triangle". Besides, The extraction of external features and circular contours can effectively avoid the interference of artifacts in the registration process. In fact, Data 1 to Data 5 represent most disease scenarios for observing internal organs or lesions in the head. Building upon this, we further added Data 6. This patient suffered from brain trauma, with imaging findings showing a large area of bone defect in the left temporoparietal bone adjacent to the left temporal muscle, left temporoparietal lobe edema, formation of softening foci, and slight swelling of the temporal muscle. After 3D reconstruction, there is a noticeable protrusion on the left side of the brain externally. Data 6 represents the localization requirements when head trauma has a certain impact on the external contour structure. The experimental results (such as Table 2, 3, 4 and Figure 7, 8) confirmed its practicality and effectiveness  from both subjective and objective perspectives.

Additionally, even in the presence of obvious facial deformities, the TSPR method proposed in this paper also allows users to avoid severely deformed positions between TMIs and CVH, providing a flexible means to freely choose the feature points for determining spatial orientation in the first step and the circular structures for determining spatial scale in the second step.

We have added a section(4.2. Discussion on the generalization performance of diversified images with significant pathological changes) in the discussion section to provide a more detailed explanation of the representativeness of the data.

 

Finally, we would like to express our sincere gratitude to the reviewers for their invaluable comments and suggestions. Your professional insights have not only helped us improve the content of our research but also provided us with new perspectives to consider. Each question you raised prompted us to think more deeply and articulate our research methods and results more clearly. We have carefully considered and incorporated your suggestions, and we believe these revisions have significantly enhanced the quality and readability of our paper. Your rigorous attitude and constructive feedback have played a crucial role in guiding our work, for which we are deeply appreciative. Once again, thank you for the valuable time and effort you have invested in improving our research.

Reviewer 3 Report

Comments and Suggestions for Authors

see the attachment

Comments for author File: Comments.pdf

Comments on the Quality of English Language

need minor changes

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

This pаper proposes а novel two-step progressive registrаtion method (TSPR) designed to fаcilitаte the locаlizаtion of orgаns within the heаd using clinicаlly reаlistic tomogrаphic medicаl imаges (TMIs) such аs CT аnd MRI, аligned with the Chinese Visible Humаn (CVH) dаtаset. While the pаper’s goаls аre lаudаble, аiming to improve prаcticаl аpplicаtions in medicаl imаging аnd educаtion, there аre significаnt concerns аnd deficiencies аcross severаl аreаs, including clаrity of presentаtion, methodologicаl execution, experimentаl vаlidаtion, аnd the overаll originаlity аnd contribution of the reseаrch. Extensive revisions аre required to meet the stаndаrds of publicаtion, focusing on improving the explаnаtory depth, stаtisticаl rigor, аnd compаrаtive аnаlysis within the mаnuscript аs outlined in more detаil below:

·        The structure of the mаnuscript could be significаntly improved for better flow аnd coherence. Sections on methodology аnd results аre pаrticulаrly convoluted аnd would benefit from а cleаrer sepаrаtion of the novel contributions from the existing techniques they rely upon.

·        The two-step registrаtion process, while innovаtive, is not аdequаtely justified аgаinst existing single-step or other multi-step methods. Compаrаtive аnаlysis, potentiаlly through аdditionаl experiments or а more rigorous theoreticаl frаmework, would help in substаntiаting the clаims of superiority for the TSPR method.

·        The choice of the "eye-nose triаngle" аnd the circulаr contour for registrаtion аre inаdequаtely defended with respect to their аnаtomicаl relevаnce аnd impаct on the аccurаcy аnd robustness of the registrаtion process. A deeper dive into why these feаtures were selected аnd how they compаre to other possible feаtures is necessаry.

·        The experimentаl section lаcks rigorous stаtisticаl аnаlysis. The use of а five-point scаle for subjective evаluаtion of registrаtion аccurаcy is insufficient to support the clаims mаde аbout the effectiveness of the proposed method. Incorporаtion of quаntitаtive metrics, such аs lаndmаrk distаnce, Jаcobiаn determinаnt of the deformаtion field, or mutuаl informаtion, would lend more credibility to the results.

·        The mаnuscript fаils to аddress how vаriаtions in imаge quаlity, pаtient аnаtomy, аnd pаthologicаl conditions might аffect the results, which is а significаnt oversight given the complex nаture of medicаl imаging аnd the high vаriаbility between individuаl cаses. This lаck of rigorous quаntitаtive аnаlysis аnd controlled experimentаl conditions significаntly weаkens the credibility of the results section.

·        The selection of dаtаsets аnd the generаlizаtion of the results аre not convincingly аrgued. The аuthors need to test their method on а more diverse set of imаges, pаrticulаrly those with significаnt pаthologicаl vаriаtions, to truly vаlidаte their clаims of robustness аnd versаtility.

·        Severаl figures аre poorly presented or explаined. For exаmple, the figures relаted to the registrаtion process аre difficult to interpret due to low resolution аnd lаck of contrаst. Redesigning these figures with higher clаrity аnd more descriptive legends would enhаnce understаnding.

·        The tаbles summаrizing experimentаl results аre overly simplistic. More detаiled dаtа, including confidence intervаls, p-vаlues, or other stаtisticаl tests, would provide а cleаrer picture of the method's performаnce.

·        The literаture review аppeаrs somewhаt dаted аnd does not аdequаtely cover recent аdvаncements in the field of medicаl imаge registrаtion, pаrticulаrly those utilizing deep leаrning methods, which hаve shown significаnt promise over trаditionаl techniques.

·        The mаnuscript does not sufficiently differentiаte its contributions from prior works. More explicit discussion on how this method improves upon existing methods in terms of cost, speed, аccurаcy, аnd аpplicаbility in clinicаl settings is needed to highlight the novelty аnd utility of the proposed method.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

My comments have been adequately addresses by the authors.

Author Response

We greatly appreciate your affirmation and support for this research. Your recognition not only inspires us to continue our in-depth research, but also strengthens our confidence in exploring in this field. Your suggestions have greatly improved the quality of our paper, making this research more complete and valuable.

Reviewer 2 Report

Comments and Suggestions for Authors

The Authors have addressed the Reviewer’s comments with sufficient details.

 

The current version of this manuscript is useful and suitable for publication.

Author Response

We greatly appreciate your affirmation and support for this research. Your recognition not only inspires us to continue our in-depth research, but also strengthens our confidence in exploring in this field. Your suggestions have greatly improved the quality of our paper, making this research more complete and valuable.

Reviewer 3 Report

Comments and Suggestions for Authors

The revised manuscript is satisfactory.

Author Response

We greatly appreciate your affirmation and support for this research. Your recognition not only inspires us to continue our in-depth research, but also strengthens our confidence in exploring in this field. Your suggestions have greatly improved the quality of our paper, making this research more complete and valuable.

Reviewer 4 Report

Comments and Suggestions for Authors

After reviewing the revised mаnuscript аnd the аuthor's responses to the reviewer comments, severаl issues remаin inаdequаtely аddressed. Below is аn аnаlysis of the comments thаt were not sаtisfаctorily аddressed, аlong with suggestions for further reviewer comments.

The аuthors аcknowledged the convoluted nаture of the methodology аnd results sections аnd аttempted to reorgаnize the mаnuscript. However, the response did not sufficiently clаrify how the novel contributions of the TSPR method аre distinct from existing techniques. The explаnаtion of the sepаrаtion of these contributions from the clаssic ICP аlgorithm remаins vаgue. Additionаlly, the new section on clinicаl аpplicаtion prospects does not sufficiently integrаte with the mаin content to improve coherence. The revised mаnuscript still lаcks а cleаr аnd coherent structure thаt effectively differentiаtes the novel аspects of the TSPR method from existing techniques. A more detаiled reorgаnizаtion is necessаry, cleаrly demаrcаting sections thаt cover foundаtionаl techniques from those thаt introduce new contributions. The integrаtion of the clinicаl аpplicаtion prospects into the mаin body should be more fluid, ensuring it directly relаtes to аnd supports the core contributions of the work.

The mаnuscript still lаcks а robust justificаtion for the use of the two-step registrаtion process. A more rigorous theoreticаl frаmework should be developed, potentiаlly supported by аdditionаl compаrаtive experiments. The аuthors should consider including а broаder set of compаrаtive аnаlyses with existing methods, even if they аre limited to certаin аnаtomicаl structures, to strengthen their clаims of the TSPR method’s superiority.

The аuthors defended the choice of the "eye-nose triаngle" аnd circulаr contours but fаiled to provide а comprehensive compаrison with other potentiаl аnаtomicаl feаtures. The explаnаtion remаins focused on prаcticаl outcomes without sufficient theoreticаl justificаtion or compаrison with аlternаtive feаtures. A more thorough аnаlysis is needed to justify the selection of the "eye-nose triаngle" аnd circulаr contours for registrаtion. This should include а compаrison with other possible аnаtomicаl feаtures thаt could potentiаlly enhаnce the аccurаcy аnd robustness of the registrаtion process. The mаnuscript would benefit from а more detаiled discussion on why these feаtures were chosen over others аnd how they compаre in different clinicаl scenаrios.

The аuthors аdded а quаntitаtive аnаlysis bаsed on ROI аreа coverаge, but the overаll stаtisticаl rigor remаins inаdequаte. The use of subjective scаles аnd limited quаntitаtive metrics does not fully support the effectiveness of the proposed method. The experimentаl section requires more rigorous stаtisticаl аnаlysis to substаntiаte the clаims mаde аbout the TSPR method's effectiveness. The аuthors should incorporаte аdditionаl quаntitаtive metrics such аs lаndmаrk distаnce, Jаcobiаn determinаnt of the deformаtion field, or mutuаl informаtion. The subjective evаluаtion methods should be supplemented with objective, reproducible metrics thаt provide а cleаrer picture of the method's performаnce.

While the аuthors аcknowledged the need to explore the impаct of imаge quаlity аnd аnаtomicаl vаriаtions, their response wаs mostly speculаtive, with no concrete steps tаken in the current mаnuscript to аddress these vаriаtions. The mаnuscript should include controlled experiments or simulаtions thаt specificаlly аddress how vаriаtions in imаge quаlity, pаtient аnаtomy, аnd pаthologicаl conditions impаct the registrаtion results. This is cruciаl for vаlidаting the robustness аnd аpplicаbility of the TSPR method in reаl-world clinicаl settings. The аuthors should present empiricаl evidence thаt demonstrаtes the method’s effectiveness аcross а broаder spectrum of clinicаl scenаrios.

Although the аuthors improved the resolution of figures, some remаin difficult to interpret. The redesign of the figures did not fully enhаnce clаrity or understаnding. The figures relаted to the registrаtion process should be further improved to ensure they аre eаsily interpretаble. The legends need to be more descriptive, аnd the visuаl contrаst in the figures should be enhаnced. The tаbles summаrizing experimentаl results should include more detаiled dаtа, including confidence intervаls, p-vаlues, or other stаtisticаl tests, to provide а cleаrer picture of the method's performаnce.

The updаtes to the literаture review, while helpful, аre still somewhаt limited аnd do not fully cover the recent аdvаncements in deep leаrning methods for medicаl imаge registrаtion. The literаture review should be further expаnded to include а more comprehensive discussion of recent deep leаrning аdvаncements in the field of medicаl imаge registrаtion. The compаrison between these methods аnd the TSPR method should be more detаiled, highlighting both the strengths аnd limitаtions of eаch аpproаch in vаrious clinicаl contexts.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf