Enhancing Thyroid Nodule Detection in Ultrasound Images: A Novel YOLOv8 Architecture with a C2fA Module and Optimized Loss Functions
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsComments:
(1) Authors mentioned the following fact regarding “Removal of Non-essential Information” under section 2.1.2 Data preprocessing: The colorbar, parameter information or any other text in the main image area imposes noise in the database and affect the model’s performance. Do authors believe this by chance, or they have any experimental proof or any previous reference from any other study? Kindly support your rationale with sufficient proof.
(2) Data augmentation is definitely an proven way to enhance the model’s performance. However, in case of medical images not all types of augmentation are permitted. Kindly refer to the augmentation styles in medical image segmentation and classification.
(3) Authors are requested to include an image of all types of augmentation performed on a single image so that it is easily understood by the audience.
“Specific augmentation methods include: horizontal and vertical flipping to increase sample diversity; cropping operations to simulate different shooting distances and perspective changes; rotation to enhance the model’s robustness to different angles; shear transformations to simulate geometric distortions in the images; brightness adjustments to enable the model to adapt to target detection under varying lighting conditions”
(4) There are many studies involving semantic segmentation of Tyroid Nodules, so why a localization/bonding box detection method such as YoLO is superior than any other semantic segmentation method? Present the rationale in discussion section and include in your objectives.
(5) A recall (True Positive Rate or successful detection of positive class) of 41.6% is extremely low. Authors should reconsider the training the model or improve the database to improve the results.
(6) In relation with above question an ROC curve would be useful to present. However given 41.6% recall value the AUC under the ROC curve would be definitely below 50%. Therefore the results are not clinically acceptable in this case.
(7) Figure 9 should be updated with the Class name and detection accuracy in a way such that it clearly identifies the class (2,4a, 4b etc) and the detection accuracies. Both numbers for class and accuracy are mixed and not conveying intended information. Also update the meaning of bounding box parameters in the caption.
(8) What is the leading parameter according to the authors which is prime in this study and vital for the Thyroid nodule detection and its value must be presented in the conclusion section.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for Authors1. Please provide detailed illustrations of the C2fA module design.
2. Please explain the results of verifying the model's generalization ability across diverse datasets.
3. Simplify the mathematical formula and provide examples to illustrate the importance of the loss function.
4. Please elaborate on the future directions for improvement and potential applications of YOLO-Thyroid.
5. Please cite more recent references from Technologies to show the relevance of your study for the journal.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsDear authors.
I have reviewed the manuscript with great interest.
However, I have some concerns.
The manuscript is too extensive, and that precludes from a gentle reading.
Introduction should clearly state what is the current problem in the diagnosis of thyroid nodules diagnosis, what has been previously investigated and how are the authors trying to give such a response to the gap of knowledge.
Material and methods is to complex and extensive also. I would invite authors to summarize to what they consider unmissable, and the rest just attach as supplementary material.
If we are talking of a new diagnosis model, a comparisson with the gold standard should be done, as well as the presentation of basic variables in diagnosis, such as accuracy, sensibility, specificity, possitive and negative predictive values.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsAuthors have answered the review questions satisfactorily. However, they are advised to improve the recall values in future models, as its value resonate with other parameters.
Author Response
Thank you for your valuable feedback. We are pleased that the responses to the review questions were satisfactory and appreciate your suggestion to improve recall values in future models. In our future work, we plan to incorporate advanced models or technologies, such as the MAMBA and KAN architectures, to further enhance the model's performance. Your insightful recommendations are greatly appreciated.
Reviewer 2 Report
Comments and Suggestions for AuthorsAccept.
Author Response
Thank you for your positive feedback and for accepting our work. Your support is greatly appreciated, and we are grateful for the opportunity to contribute to this field.
Reviewer 3 Report
Comments and Suggestions for AuthorsI have no more comments.
Thank you for the effort.
Author Response
Thank you for your kind feedback and for recognizing our efforts. We truly appreciate your time and consideration in reviewing our work.
