M3ASD: Integrating Multi-Atlas and Multi-Center Data via Multi-View Low-Rank Graph Structure Learning for Autism Spectrum Disorder Diagnosis
Round 1
Reviewer 1 Report
Comments and Suggestions for Authors
In attached PDF file.
Comments for author File: 
Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for Authors
The paper presents interesting research ideas. However, the paper needs some improvements and additions, as indicated in the comments below.
1. How do the authors plan to address the potential limitations of rs-fMRI alone for ASD diagnosis by integrating multimodal or electronic sensor-based data?
2. Could the authors elaborate on how the model can be generalized to structural imaging or hybrid biomedical devices?
3. Have the authors considered deep learning models based on time series in addition to graphical representations for ASD prediction?
4. What is the translational potential of the model for rehabilitation and patient monitoring contexts? In this regard, the authors should supplement the paper with recent studies. I am not asking the authors to implement a model from scratch, but only to integrate the study doi: 10.3390/electronics14112268, which provides a valuable interdisciplinary link between computational models and patient-specific systems.
5. Could the proposed approach work with sensor-based systems for precision diagnostics?
6. Could the authors elaborate on how the model addresses underlying or hidden connectivity patterns?
7. Would the method be adaptable to other biomedical disorders besides ASD?
8. How do the authors plan to adapt M3ASD to patient-specific monitoring devices?
9. Could the model be extended to real-time monitoring in a clinical setting?
10. Why did the authors not consider fuzzy modeling to address uncertainty in ASD diagnosis?
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for Authors
The paper proposes the M3ASD method, a multi-view low-rank subspace graph structure learning method for integrating multi-atlas and multi-center data for automated ASD (autism spectrum disorder) diagnosis.
The following comments are made regarding the article.
- The text in Figure 1 appears blurry. This comment also applies to other figures.
- In formula (2), it is necessary to justify the choice of alpha values (0.6, 0.7, and 0.2). Why is the number of views equal to 3 (line 232)?
- What does φ mean in formula (2)? Obviously, φ must have ij indices.
- Why does Figure 3 refer to “Input GSL”?
- It is necessary to justify the choice of values for all hyperparameters (including λ1 and λ2).
- What is the upper limit of summation in formula (4)?
- What does P mean in formula (5)?
- What is the difference between σi (formula (4)) and σ (formula (11))? I recommend using different notations for different parameters.
- Use the same writing style for variables in the text and in formulas, for example, italics (e.g., for ACC, SEN, SPE, TP, TN, FP, and FN).
- In Figure 6, it is appropriate to use bar charts since the number of views is discrete.
- It is necessary to apply some statistical criteria to prove the advantage of the proposed method.
- What are the limitations of the applicability of the proposed method?
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments and Suggestions for Authors
I would like to thank the authors for their contribution to the research. In order to improve the proposed document, the authors are required to supplement it and respond to the following comments.
1. Could the authors extend the M3ASD to physiological data based on electronic sensors with rs-fMRI for a multimodal diagnostic picture of ASD?
2. Could the integration of temporal learning models, such as LSTM or U-Net architectures, improve ASD pattern recognition in longitudinal datasets?
3. Could the M3ASD model be adapted for continuous patient monitoring or neurorehabilitation systems?
4. Have the authors considered integrating optical or photon-based sensors into ASD neuroimaging systems to improve signal resolution? The recent study doi:10.3390/s24175568
shows how advanced photonic sensors enable real-time biomedical detection.
5. Is the proposed system adaptable to other disorders characterised by hidden functional or structural abnormalities?
6. Can the M3ASD model be integrated with electromyographic or electrophysiological data for patient-specific diagnostics?
7. Would the system support real-time clinical decision-making through integrated or portable systems?
8. How could uncertainty quantification be addressed to improve diagnostic reliability in heterogeneous patient populations?
9. Is the model capable of handling domain changes between non-medical datasets? Authors should supplement the Discussion with the study doi: 10.1109/SAS60918.2024.10636531,
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for Authors
The revised manuscript has been substantially improved and most reviewer comments have been adequately addressed. The addition of statistical significance testing, enhanced neuroscientific interpretation, comprehensive hyperparameter analysis, and expanded discussion of limitations significantly strengthens the work. However, 02 critical issues must be resolved before acceptance as follows:
* Code Availability: The GitHub repository (https://github.com/shuoyang031102/) should be formally cited in the Data Availability Statement (lines 649-651). More importantly, to ensure the work is reproducible, the repository must be updated to include the complete implementation code for the M³ASD framework—including all preprocessing scripts, model architectures, comments, and evaluation pipelines—along with a detailed README file explaining how to replicate all experimental results.
* Computational Complexity Analysis: Reviewer comment 7 explicitly requested computational complexity analysis. Authors stated it would be "addressed this in future work" without adding any analysis to the manuscript. This is unacceptable for a methods paper. Minimum required content should include some analyses about training efficiency, inference efficiency, memory requirements, complexity analysis, scalability etc.
Other recommended revisions:
- Enhanced Ablation Study (Table 3): the response to Comment 10 states: "We also tested removing combinations of modules (e.g., both multi-view and low-rank), which led to a more severe performance drop than the sum of individual removals, indicating a synergistic effect." but the data is not shown in Table 3. Only single-module ablations are presented.
- Figure 8 is a key interpretability result but lacks quantitative validation and some visual elements. Adding statistical validation makes it more rigorous and convincing.
- Reference Formatting: Reference 1 still noted as having "multiple mixed citations and incorrectly formatted"; it has to be fixed completely. Add DOIs to references where available (check references 3, 4, 11, 13, 21, 29, 31, 41).
Conclusion: The manuscript is substantially improved and scientifically considerable. The core contributions are valuable and the method is well-executed within its scope. Two critical issues (code availability documentation, computational complexity analysis) have to be solved due to their essential methodological importance.
Recommendation: MINOR REVISION
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for Authors
I thank the authors for their replies to the comments. I have no further comments to make.
Author Response
We sincerely appreciate the reviewer's comprehensive evaluation and positive reception of our manuscript. The lack of specific critiques in your feedback serves as a strong endorsement of the research quality, methodological rigor, and clarity of presentation, which we highly value. Your expertise and meticulous review process have not only validated the robustness of our proposed M³ASD framework but also reinforced the significance of our contributions to the field of neuroimaging-based autism diagnosis. The absence of raised concerns indicates that the integration of multi-atlas and multi-center data, along with the novel graph structure learning approach, was effectively communicated and substantiated. This affirmation encourages us to persist in our efforts to refine and expand this work, particularly in exploring real-time clinical applications and multi-modal data fusion. We remain dedicated to upholding the highest standards of scientific inquiry and look forward to continued advancements in this critical research domain. Once again, we thank you for your time and insightful assessment, which have been instrumental in bolstering our confidence and commitment to excellence.
Reviewer 3 Report
Comments and Suggestions for Authors
The article may be published in the present form.
Author Response
We sincerely appreciate the reviewer's comprehensive evaluation and positive reception of our manuscript. The lack of specific critiques in your feedback serves as a strong endorsement of the research quality, methodological rigor, and clarity of presentation, which we highly value. Your expertise and meticulous review process have not only validated the robustness of our proposed M³ASD framework but also reinforced the significance of our contributions to the field of neuroimaging-based autism diagnosis. The absence of raised concerns indicates that the integration of multi-atlas and multi-center data, along with the novel graph structure learning approach, was effectively communicated and substantiated. This affirmation encourages us to persist in our efforts to refine and expand this work, particularly in exploring real-time clinical applications and multi-modal data fusion. We remain dedicated to upholding the highest standards of scientific inquiry and look forward to continued advancements in this critical research domain. Once again, we thank you for your time and insightful assessment, which have been instrumental in bolstering our confidence and commitment to excellence.
Reviewer 4 Report
Comments and Suggestions for Authors
I have no further comments or observations to add.
Author Response
We sincerely appreciate the reviewer's comprehensive evaluation and positive reception of our manuscript. The lack of specific critiques in your feedback serves as a strong endorsement of the research quality, methodological rigor, and clarity of presentation, which we highly value. Your expertise and meticulous review process have not only validated the robustness of our proposed M³ASD framework but also reinforced the significance of our contributions to the field of neuroimaging-based autism diagnosis. The absence of raised concerns indicates that the integration of multi-atlas and multi-center data, along with the novel graph structure learning approach, was effectively communicated and substantiated. This affirmation encourages us to persist in our efforts to refine and expand this work, particularly in exploring real-time clinical applications and multi-modal data fusion. We remain dedicated to upholding the highest standards of scientific inquiry and look forward to continued advancements in this critical research domain. Once again, we thank you for your time and insightful assessment, which have been instrumental in bolstering our confidence and commitment to excellence.
