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.
Thanks to the authors for their efforts to improve the work. This version addressed the concerns of the reviewers successfully. It can be accepted. Congrats!.
Thank you very much for addressing all my concerns. I appreciate your gentle responses to each of my comments.
Authors provided more information on how the features used to train the model were calculated.
Authors described in detail the methods used in the comparison.
Please review the adjective infected in the following sentence of the Section 4.3:
"The digital image resolution is 1500×1000 pixels of which 76 are infected with DR.."
DR is not an infectious disease.
**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.
**PeerJ Staff Note:** It is PeerJ policy that additional references suggested during the peer-review process should only be included if the authors agree that they are relevant and useful.
**Language Note:** The review process has identified that the English language must be improved. PeerJ can provide language editing services - please contact us at [email protected] for pricing (be sure to provide your manuscript number and title). Alternatively, you should make your own arrangements to improve the language quality and provide details in your response letter. – PeerJ Staff
This paper focuses on improving Diabetic Retinopathy (DR) detection in fundus images by introducing a Graded Region-of-Interest (ROI) Feature Dissection method. The objective is to enhance the identification of micro-level retinal changes for more precise DR detection. The proposed method first identifies the ROI based on maximally differentiated features. Then, feature changes are iteratively validated using a recurrent neural network with a cross-pooling function applied to the edges of the ROI features. The method was evaluated on a publicly available fundus image dataset from the Kaggle repository.
The paper needs a revision to correct some details:
1) Please use italics for equations and variables, for example, n x m.
2) Please ensure that all terms are correctly defined, for example, c_a in Eq. 2.
3) I think that the next expression is not accurate: “of which 76 are infected with DR”. It is because DR is not an infectious disease.
4) Please provide more detailed captions for all Figures and Tables. This will help the reader better understand.
5) Please highlight the best results in Tables 3 and 4, for their better identification.
1) Please provide more information on how the features used to train the model were calculated. This is because the comparative analysis mentions that the features and edges were varied.
2) Can you include the definition of the metrics used to evaluate the performance of the DR detection?
3) Please include the definitions of the metrics used to evaluate the performance of the DR detection method.
4) Did you implement the comparison methods yourself, or did you use existing implementations? Was the detection delay measured using the same computer for all methods?
5) Please provide more details of the dataset used, for example, the number of classes and the number of images per class.
1) Can you describe in detail the methods used in the comparison: RTNet (Huang et al., 2022), MCNN-UNet (Skouta et al., 2022), and DRFEC (Das et al., 2023). Did these methods use the same dataset?
2) In the comparative analysis section, it would be welcome to include a discussion of the significant differences between the proposed method and the related works used for this purpose.
1) The work is interesting and the results are sound. However, the description of the proposed method is difficult to follow. I respectfully suggest revising Section 3 to improve clarity and make the methodology easier to understand. It may help to first describe the RNN architecture, followed by the proposed methodology.
1. Language needs major editing for grammar and clarity. Terms like “purses” and “violating changes” are unclear.
2. Equations use non-standard symbols (e.g., “7” for minus) and need proper formatting and variable definitions.
3. Figures are referenced but lack detailed captions and in-text explanations.
4. The introduction does not clearly define the problem gap or articulate the novelty.
5. The related works section is descriptive, not analytical, and there is no clear contrast with the proposed method.
6. Formatting inconsistencies throughout nspacing, notation, and symbol use.
7. References are recent but not well integrated into the discussion.
8. No raw data or code shared, reproducibility is limited.
1. The research objective is relevant, but the experimental design lacks justification for using RNN over other architectures like CNN or Transformers.
2. Dataset details are minimal—no mention of class balance, preprocessing steps, or annotation quality.
3. Evaluation is done on a limited dataset (100 training, 76 testing images), which is insufficient for generalization—needs cross-validation or additional datasets.
4. No ablation study is provided to isolate the impact of the cross-pooling mechanism.
5. Hyperparameter tuning lacks methodology—values are given without explanation of how they were selected.
6. MATLAB implementation on low-resource hardware may limit scalability—needs justification.
7. Data augmentation or regularization methods are not discussed.
1. Reported performance gains are promising, but no statistical tests (e.g., p-values, confidence intervals) are provided to validate significance.
2. No qualitative visual results (e.g., segmented images) are shown to support the numerical findings.
3. Comparison with baseline models is limited to a few methods—broader benchmarking is recommended.
4. The RNN-based approach lacks justification for effectiveness over CNN-based detectors commonly used in medical imaging.
5. Error analysis is shallow no discussion of failure cases or model limitations.
6. Claims about improvement in precision and specificity need a clearer linkage to architectural contributions.
1. The manuscript would benefit from citing more recent literature (2023–2024) on diabetic retinopathy detection, especially works involving advanced transformer models, attention mechanisms, or hybrid deep learning approaches.
1. Wang, W., Yuan, X., Wu, X., & Liu, Y. (2017). Fast Image Dehazing Method Based on Linear Transformation. IEEE Transactions on Multimedia, 19(6), 1142-1155. doi: 10.1109/TMM.2017.2652069
2. Tian, J., Zhou, Y., Chen, X., AlQahtani, S. A., Chen, H., Yang, B.,... Zheng, W. (2024). A Novel Self-Supervised Learning Network for Binocular Disparity Estimation. CMES - Computer Modeling in Engineering and Sciences, 142(1), 209-229. doi: https://doi.org/10.32604/cmes.2024.057032
3. Wu, Z., Sun, W., & Wang, C. (2024). Clinical characteristics, treatment, and outcomes of pembrolizumab-induced uveitis. Investigational New Drugs, 42(5), 510-517. doi: 10.1007/s10637-024-01464-w
4. Liang, J., He, Y., Huang, C., Ji, F., Zhou, X.,... Yin, Y. (2024). The Regulation of Selenoproteins in Diabetes: A New Way to Treat Diabetes. Current Pharmaceutical Design, 30(20), 1541-1547. doi: https://doi.org/10.2174/0113816128302667240422110226
5. Song, W., Wang, X., Guo, Y., Li, S., Xia, B.,... Hao, A. (2024). CenterFormer: A Novel Cluster Center Enhanced Transformer for Unconstrained Dental Plaque Segmentation. IEEE Transactions on Multimedia, 26, 10965-10978. doi: 10.1109/TMM.2024.3428349
6. Sun, J., Yu, X., Li, W., Jia, B., Shi, D., Song, Y.,... Jiang, C. (2025). Real-time accurate detection and analysis of breath acetone using CRDS: Toward metabolic dynamic monitoring and potential application. Sensors and Actuators B: Chemical, 433, 137422. doi: https://doi.org/10.1016/j.snb.2025.137422
7. Wang, G., Ma, Q., Li, Y., Mao, K., Xu, L.,... Zhao, Y. (2025). A skin lesion segmentation network with edge and body fusion. Applied Soft Computing, 170, 112683. doi: https://doi.org/10.1016/j.asoc.2024.112683
8. Hu, T., Meng, S., Liu, C., Fang, W., Xia, Z., Hu, Y.,... Xia, X. (2025). LCN2 deficiency mitigates the neuroinflammatory damage following acute glaucoma. Theranostics, 15(7), 2967-2990. doi: 10.7150/thno.104752
9. Zhang, N., Wang, Y., Li, W., Wang, Y., Zhang, H., Xu, D.,... Tang, H. (2025). Association between serum vitamin D level and cardiovascular disease in Chinese patients with type 2 diabetes mellitus: a cross-sectional study. Scientific Reports, 15(1), 6454. doi: 10.1038/s41598-025-90785-8
**PeerJ Staff Note:** The PeerJ's policy is that any additional references suggested during peer review should only be included if the authors find them relevant and useful.
2. Several concepts introduced (e.g., cross-pooling violation, memory state in RNN) need clearer explanation with references to related existing methods.
3. The title and abstract should be rephrased to reflect the methodological novelty and main contributions better.
4. Consider simplifying the mathematical formulations and ensuring they are interpretable to a broader audience.
5. The conclusion section should better emphasize limitations and provide a clearer roadmap for future improvements.
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.