BUU-LSPINE: A Thai Open Lumbar Spine Dataset for Spondylolisthesis Detection

Round 1

Reviewer 1 Report

1. The authors should follow IMRAD structure in writing the abstract.

2. A brief introduction on spondylolisthesis should be put in the abstract.

3. Deep learning method should be explained in the abstract.

4. Detection result statistics should be put in the abstract.

5. Discussion of the implication of the obtained result on spondylolisthesis treatment and prevention should be added in the abstract and conclusion.

6. A separate section of the methodology should be added to explain the architecture of the DL method for this spondylolisthesis detection.

4. 

Author Response

Thank you for taking the time to review our work. We truly value your effort and feedback, as it is crucial for improving our research results. Your insights have been invaluable in identifying previously overlooked weaknesses. We have diligently incorporated your valuable advice and comments into our paper. Please allow us to outline our updates to ensure you understand our revisions. Once again, we sincerely appreciate your helpful comments. Thank you.

Comments and Suggestions for Authors

1. The authors should follow IMRAD structure in writing the abstract. ☒

We revised the manuscript and updated the abstract structure to follow IMRAD, as shown in lines 13 – 33.

2. A brief introduction to spondylolisthesis should be put in the abstract. ☒

We included a brief introduction about spondylolisthesis in the abstract, as shown in lines 14 – 17.

3. Deep learning methods should be explained in the abstract. ☒

We added a brief explanation of deep learning methods and our approach to predict spondylolisthesis in the abstract, as shown in lines 21 – 28.

4. Detection result statistics should be put in the abstract. ☒

We included the result statistics of three experiments in the abstract, as shown in lines 14 – 28.

5. Discussion of the implication of the obtained result on spondylolisthesis treatment and prevention should be added in the abstract and conclusion. ☒

We included the spondylolisthesis prediction idea in the abstract and conclusion, as shown in lines 14 – 17 and 400 – 403.

6. A separate section of the methodology should be added to explain the architecture of the DL method for this spondylolisthesis detection. ☒

We added separate sections to describe the main ideas of our spondylolisthesis prediction, as shown in section 2.4.3 between lines 262 - 307.

Author Response File: Author Response.pdf

Reviewer 2 Report

Minor revision is suggested

1.      More lumbar detection and corner extraction results should be presented in Figure 7 and Figure 9. Besides, Figure 7 should be replaced with a clearer picture.

2.      The authors only show average accuracy for IoU thresholds of 0.5 to 0.95. In fact, the IoU thresholds of 0.5, 0.75, and 0.95 respectively should also be shown.

3.      The authors selected different network models for comparison in three applications (corner extraction, lumbar detection, and prediction of lumbar spondylolisthesis). Why choose so many different models?

4.      The authors used only three methods for comparison in three applications. More methods of comparison need to be added.

The English should be further polished

Author Response

Thank you for taking the time to review our work. We truly value your effort and feedback, as it is crucial for improving our research results. Your insights have been invaluable in identifying previously overlooked weaknesses. We have diligently incorporated your valuable advice and comments into our paper. Please allow us to outline our updates to ensure you understand our revisions. Once again, we sincerely appreciate your helpful comments. Thank you.

Comments and Suggestions for Authors

Minor revision is suggested.

1. More lumbar detection and corner extraction results should be presented in Figure 7 and Figure 9. Besides, Figure 7 should be replaced with a clearer picture. ☒

We fixed the quality of figure 7, as shown in lines 242 – 243 and added more results of lumbar detection and corner extraction, as shown in section 3.1 – 3.2 and lines 309 – 358.

2. The authors only show average accuracy for IoU thresholds of 0.5 to 0.95. In fact, the IoU thresholds of 0.5, 0.75, and 0.95 respectively should also be shown. ☒

We included the mean average precision of the IoU thresholds of 0.5 and 0.75, as shown in Table 8 as shown in lines 315.

3. The authors selected different network models for comparison in three applications (corner extraction, lumbar detection, and prediction of lumbar spondylolisthesis). Why choose so many different models? ☒

We utilized multiple network models in our study to gain a more comprehensive understanding of their performances, strengths, and limitations, guiding us toward better evaluating the potential of our BUU-LSPINE dataset.

4. The authors used only three methods for comparison in three applications. More methods of comparison need to be added. ☒

We conducted more methods in lumbar vertebrae detection and vertebral corner point extraction, as shown in sections 3.1 and 3.2 between lines 309 to 358.

Author Response File: Author Response.pdf

Reviewer 3 Report

Dear Researcher,

thank you for your interesting work that seems to perfectly match my expertise area in spine desease.
The paper apperas weel organized and presented. However, I propose some suggestions and questions:

- it would be appropriate to highlight from the beginning of the work the meaning of the various acronyms that appear in the paper

- could be very intresting to know algorithmic scaffoldinf for point detecion

- it would be appropiate to know the Diagnostic parameters for RX rays. As you know, the quality of immaging diagnostic and the contrast of pixels, directly affects the HOUNSFIELD scale for tissue segmentation

- the number of trial experiments must be improved

sure of your evaluation, I cordially greet you

Author Response

Thank you for taking the time to review our work. We truly value your effort and feedback, as it is crucial for improving our research results. Your insights have been invaluable in identifying previously overlooked weaknesses. We have diligently incorporated your valuable advice and comments into our paper. Please allow us to outline our updates to ensure you understand our revisions. Once again, we sincerely appreciate your helpful comments. Thank you.

Comments and Suggestions for Authors

Dear Researcher,

Thank you for your interesting work that seems to perfectly match my expertise area in spine disease.

The paper appears well organized and presented. However, I propose some suggestions and questions:

Point 1: It would be appropriate to highlight from the beginning of the work the meaning of the various acronyms that appear in the paper. ☒

Response 1: We added the meaning of acronyms in the beginning of our study, as shown in lines 34 – 39.

 Point 2: It could be very interesting to know algorithmic scaffolding for point detection. ☒

Response 2: We added more information about the vertebral corner points detection experiment in the sections 2.4.2 and 3.2 between lines 244 – 261 and 333 – 358.

 Point 3: It would be appropriate to know the Diagnostic parameters for RX rays. As you know, the quality of imaging diagnostic and the contrast of pixels directly affects the HOUNSFIELD scale for tissue segmentation. ☐

Response 3: We have included the X-ray imaging parameters (kV and mAs) used to capture the plain film images in our dataset. These parameters are detailed in lines 120 to 127 and row 18 of Table 2.

 Point 4: The number of trial experiments must be improved. ☐

Response 4: We conducted more trail experiments in lumbar vertebrae detection and vertebral corner point extraction, as shown in sections 3.1 and 3.2 between lines 309 to 358.

sure, of your evaluation, I cordially greet you.

Author Response File: Author Response.pdf