All reviews of published articles are made public. This includes manuscript files, peer review comments, author rebuttals and revised materials. Note: This was optional for articles submitted before 13 February 2023.
Peer reviewers are encouraged (but not required) to provide their names to the authors when submitting their peer review. If they agree to provide their name, then their personal profile page will reflect a public acknowledgment that they performed a review (even if the article is rejected). If the article is accepted, then reviewers who provided their name will be associated with the article itself.
The reviewers' concerns were addressed in the revised paper. I recommend that it be accepted for publication.
[# PeerJ Staff Note - this decision was reviewed and approved by Shawn Gomez, a PeerJ Section Editor covering this Section #]
Claims regarding clinical readiness, robustness, and superiority are not supported by the data or by external validation.
Generalizability cannot be claimed.
Ignores volumetric continuity
The experimental design is not rigorous, lacks external validation
Presenting only 2D models and avoiding 3D CNN comparisons skews the fairness of results.
Reviewer comments were analyzed in detail. It is seen that the corrections made by the authors comply with the basic reporting criteria. In particular, the Introduction has been comprehensively restructured, clearly summarizing specific lung nodule detection studies in the literature (YOLO-based, Transformer-based, and segmentation-based approaches) and strongly grounding the motivation of the study.
As requested by the journal, the authors presented the contributions of the study item by item and clearly at the end of the Introduction, as requested by the reviewers, so that the study's novelty could be better understood. hale has been received.
In the article, terminology was used to make Criticism appropriate. Authors by Concepts such as CT scan, 2D slice, image, and annotation right one as defined. The article's language has been corrected.
As a result, the authors responded appropriately and adequately to all of the referee's key reporting expectations.
The reviewers correctly identified methodological shortcomings in the paper. Separation of the data and 10-fold cross-validation improved the accuracy of the results and are detailed in the paper. The dataset is also tabulated. The authors provide a comparative table of related studies. In the methodology section, the MS-YOLO11 architecture, MCA, and SMAT components are explained by giving their mathematical models.
Medical imaging domain-specific metrics such as DR, SEN, FPs/scan, and ACC were added to the study. In addition, a random number seed was fixed in all experiments, thus enhancing reproducibility.
In conclusion, the methodological arrangements adequately and appropriately addressed all experimental design expectations.
In the study, inconsistencies regarding the training time were corrected, and it was stated that experimental studies were carried out over 300 epochs. Training and loss curves are explained again. The success of MS-YOLO11 with different metrics is shown. Model SOTA is compared again under fair conditions.
The results are revisited and presented with adequate evaluations as requested by the reviewers.
The comments of the referees have been carefully considered by the authors. The text language has been simplified, and explanations have been used. Figure descriptions have been revised and the conclusion has been simplified.
The corrections made by the authors have made the manuscript more coherent and readable. The article can be acceptable as is.
**PeerJ Staff Note:** Please ensure that all review, editorial, and staff comments are addressed in a response letter and that any edits or clarifications mentioned in the letter are also inserted into the revised manuscript where appropriate.
**Language Note:** When preparing your next revision, please ensure that your manuscript is reviewed either by a colleague who is proficient in English and familiar with the subject matter, or by a professional editing service. PeerJ offers language editing services; if you are interested, you may contact us at [email protected] for pricing details. Kindly include your manuscript number and title in your inquiry. – PeerJ Staff
The manuscript introduces a novel lung nodule detection framework, MS-YOLO11, which integrates Multidimensional Collaborative Attention (MCA) and Synergistic Multi Attention Transformer (SMAT) into the YOLO11 architecture.
However, the motivation presented in the Introduction section is underdeveloped. It primarily focuses on generic limitations of object detection algorithms without discussing the specific landscape of lung nodule detection. Numerous prior works have tackled this domain effectively; these should be summarized and contrasted.
The contributions of this study should be more explicitly stated, preferably at the end of the Introduction, in a bullet-point format to enhance clarity and readability.
The dataset description is ambiguous. It is unclear whether the authors are referring "images" to the number of CT scans or 2D slices. Consistency with terminology and alignment with prior LUNA16-based studies are critical.
The Related Work section would benefit from a tabular summary comparing recent lung nodule detection methods in terms of datasets, architectures, and key performance metrics.
The Methodology section should be reorganized. The first subsection should describe the overall architecture using Figure 3, clearly explaining how the various components (YOLO11, MCA, and SMAT) interact. Only afterward should each component (MCA and SMAT) be explained in detail, with motivations for their inclusion.
The evaluation metrics currently used seem biased toward general object detection rather than lung nodule detection. Established metrics such as detection rate, sensitivity, and false positives per scan (FPs/scan) are commonly used in previous works (e.g., TMI.2019.2935553 and Eng. Appl. AI. 2023.107597), and their omission raises concerns.
The data partitioning strategy is poorly explained. Best practices with at least 5-fold cross-validation should be followed and reported to ensure robustness and reproducibility. Furthermore, the number of nodules, scans, and slices used should be clearly specified.
Figure 4 is unclear and confusing. If it represents model performance over training epochs, why does it stop at 300 epochs, while Table 1 states the model is trained for 700 epochs? The training curve appears to flatten prematurely, suggesting the model may have already converged or stopped learning further. Figure 5 suffers from a similar issue and requires clarification.
The manuscript lacks a comparison with state-of-the-art (SOTA) methods, which is critical for establishing the value of the proposed model. A comparative table with recent studies using LUNA16 or LIDC datasets should be included to contextualize the performance.
1. Reduce the length of the conclusion section and ensure figure captions and legends are self-explanatory and aligned with the text discussion.
• All of the language used in the manuscript is technically correct and written in a clear, professional manner.
• Limitations: A few lines are too complicated or thick for a diverse readership, despite being comprehensible. Take into account additional simplicity for non-specialists and clinicians.
• The methodology is thorough, covering ablation tests, training, hardware/software configuration, and data preparation.
• Limitations: • The study's use of a single publicly available dataset (LUNA16) and lack of external clinical validation restricts its capacity to demonstrate generalizability and robustness. • All studies are conducted on 2D CT slices, excluding the use of multi-slice continuity or available 3D context, which may affect clinical realism and performance on real-world multi-slice data. • No experiments are conducted on other datasets or in situations involving potential users. • Code/data sharing facilitates reproducibility; nevertheless, random seed settings, low-level library variance, and inference latency—all of which have an impact on deployment claims and repeatability—are not discussed.
• Ablation investigations and robust comparisons to baselines (YOLO series, SSD, Faster R-CNN, MSDet) are included in the experiments.
• The limitations of the model, including its 2D architecture and the requirement for more extensive multicenter testing, are acknowledged in the discussion.
•Limitations: Clinical safety interpretation is limited due to the lack of rigorous error analysis (types, reasons of misclassification), uncertainty quantification, and ensemble/postprocessing.
• In accordance with review guidelines, figure/image alteration is not specifically reviewed; adherence to appropriate image ethics should be mentioned.
• There is no human-in-the-loop validation or evaluation with radiologists; clinical interpretability and trust have not been established.
Strengths:
• Innovative MCA and SMAT design substantially improves metrics, with robust ablation and benchmark experiments.
• Data and code availability strengthen reproducibility.
Suggestions for Improvement and Limitations:
1. Generalizability: Future work should validate on additional, multi-center datasets, and/or real-world scans to rule out overfitting and show versatility.
2. 3D/clinical realism: Upgrade experiments to handle volumetric 3D data, or show how the current 2D design adapts to a clinical context where nodule continuity is critical.
3. Error analysis: Add detailed breakdowns of false positives/negatives and case studies of difficult scenarios.
4. Interpretability: Systematically analyze and explain attention maps in terms of anatomical/clinical features, and demonstrate model trust/value for radiologists.
5. Segmentation tradeoffs: More deeply study which nodule subtypes benefit from segmentation vs. unsegmented data. Are some edge cases unrecoverable?
-
-
-
This paper addresses a highly sensitive topic (Pulmonary Nodule Detection). The researchers employed advanced technologies, "YOLO11," and combined multiple approaches (Multidimensional Collaborative Attention and Synergistic Multi-Attention Transformer) to achieve more accurate results. The study was comprehensive, beginning with meticulous data preparation and image extraction for training the model. The methods were clearly explained, with an emphasis on the advantages of each technique and their impact on accuracy. Furthermore, the researchers compared several versions of YOLO, demonstrating the robustness and performance of YOLOv11. Overall, this work is well-organized, and the results obtained strongly reflect the quality and reliability of this paper.
All text and materials provided via this peer-review history page are made available under a Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.