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.
Thank you for addressing the reviewer concerns.
[# PeerJ Staff Note - this decision was reviewed and approved by Catherine Myers, a PeerJ Section Editor covering this Section #]
The overall quality of the manuscript has improved, and previously unclear sections have also been clarified. Thank you.
The overall quality of the manuscript has improved, and previously unclear sections have also been clarified. Thank you.
The overall quality of the manuscript has improved, and previously unclear sections have also been clarified. Thank you.
The overall quality of the manuscript has improved, and previously unclear sections have also been clarified. Thank you.
Dear Authors,
Thank you for the revisions and improvements you have made to your manuscript. Nevertheless, several critical concerns remain to be addressed.
In particular, the reviewers have raised methodological issues that require your attention:
You are requested to provide a clear justification for the use of your chosen imaging protocols, explicitly responding to the reviewers’ comments.
You should also discuss and clarify (and revise where necessary) your decision to omit consideration of the Site of Occlusion and the TICI Scale, as the reviewers regard this omission as methodologically inappropriate.
I expect that the revised submission will comprehensively address all the concerns raised by Reviewer 2. As these concerns are closely related to the practical healthcare aspects of your work, you are strongly encouraged to incorporate feedback from experts in the relevant domain.
The authors have satisfactorily addressed the concerns.
As mentioned earlier, there is no such experiment performed in this study. This is a retrospective study of the routine MRI images. The efforts made by the authors are sufficient to be published.
Manuscript satisfactorily improved from the previous version.
If available, authors may add a figure showing the core and penumbra of the stroke in MR images.
Please see the section 'Questions for the Authors
Please see the section 'Questions for the Authors
Please see the section 'Questions for the Authors
Thank you for your manuscript and for the revisions you have made. However, not all comments from Round 1 have been adequately addressed. In particular, the ethical aspects remain insufficiently clarified.
Furthermore, there are several major methodological issues that still need to be considered:
Site of Occlusion and TICI Scale
The location of occlusion is a critical factor, as it strongly influences collateral flow and time to reperfusion. These variables directly affect ML analysis results, and ignoring them is not methodologically correct.
Similarly, the role of the TICI score is not sufficiently analyzed. It is well known that patients with low TICI grades have poor prognosis regardless of the initial penumbra size. Failed thrombectomy in these cases is a much stronger prognostic factor than imaging parameters. Only in situations where the infarct is already predominantly core does the success of thrombectomy lose its prognostic importance – but in such cases active therapy is usually not undertaken.
Imaging Algorithm
The diagnostic pathway described in your manuscript (non-contrast CT → CTA → MRI with DWI and ADC) does not reflect generally accepted international practice and is not clearly justified.
DWI and ADC are typically applied in specific situations: when CT is negative and stroke onset is unknown, or in posterior circulation infarcts and lacunar cases. In the majority of acute ischemic stroke cases, non-contrast CT and CTA are sufficient for decision-making regarding thrombolysis and thrombectomy.
CTP is recommended only in selected, specific scenarios, not as a routine approach. The imaging sequence you describe is outside guideline-based practice and results in unnecessary delays. Considering the principle “time is brain,” such an approach is not acceptable and may negatively affect patient outcomes.
Summary
Although the manuscript represents an important effort and a valuable dataset, the lack of clarity regarding ethics, the inappropriate imaging protocol, and the omission of occlusion site and TICI scale analysis significantly undermine the reliability and clinical relevance of the findings. Without addressing these issues, the manuscript cannot be considered ready for publication.
It still remains unclear why the MRI protocol included only diffusion-weighted imaging (DWI) and apparent diffusion coefficient (ADC) sequences. These sequences alone do not adequately demonstrate the extent of ischemia. Was this protocol applied solely for research purposes? Why was the FLAIR sequence not included? This creates the impression that the imaging was performed primarily for research rather than clinical decision-making, which raises important ethical concerns about the study.
The authors seem to have contradicted themselves. The statement “research purposes, and urgent treatments such as thrombolysis or thrombectomy were not delayed” is not accurate. As already mentioned, patients underwent MRI with at least two sequences, which inevitably takes additional time. Do the authors consider MRI an examination that does not take minutes to perform?
I must emphasize once again that DWI alone cannot precisely define the ischemic core. The authors’ statement “We defined the DWI lesion as the ischemic core” is therefore not accurate and oversimplifies the pathophysiology. Additional sequences and/or perfusion studies are required for a reliable assessment.
Dear Authors,
After carefully reviewing the feedback provided by both reviewers, it is my opinion that your manuscript may benefit from a significant revision.
One of the reviews highlights the scientific relevance and potential impact of your work, particularly its integration of cerebrovascular morphology and radiomics using machine learning. The comments highlight several areas in need of clarification and refinement. These include issues with writing clarity, the use of outdated citations, and insufficient explanation of technical methods such as the "sparse representation method." Also, review highlights inconsistencies in figure and table labeling, unclear referencing within the methodology, and the absence of a clearly defined statistical analysis section.
There are also more substantial concerns. The primary issue revolves around the inadequacy and vagueness of the imaging protocol. The methods section appears to be lacking essential details regarding imaging modalities, acquisition timing, diagnostic criteria, and consistency across the patient cohort. Specifically, the reliance on DWI and ADC alone to define infarct core and penumbra is not clinically valid and is considered insufficient in the current standard of care. Additionally, while CT perfusion is mentioned at one point in the text, it is not reflected in the patient selection table, revealing a critical inconsistency in your methodology.
Additionally, there are ethical concerns associated with the imaging workflow described. The sequence of complete CT protocols followed by MRI appears to lack clinical justification and may delay urgent treatments, such as thrombectomy, in time-sensitive settings. A review raises important questions about patient safety and clinical feasibility.
It would be helpful to provide a clear explanation of how vascular features were extracted, whether this was done manually or automatically, and how these features were aligned or integrated with diffusion imaging results. The criteria for patient inclusion, the role of treatment outcomes such as TICI scores, and the rationale behind feature selection for the model construction remain underexplained.
I agree with reviewers' concerns regarding the current form of the manuscript, which does not meet the standards for methodological rigor, clarity, or ethical transparency. A thorough and substantive revision is necessary. This includes rewriting the Materials and Methods section with sufficient technical detail, providing a comprehensive and clinically realistic imaging protocol, justifying the imaging decisions from both scientific and ethical perspectives, clarifying the modeling workflow, and explicitly stating the study's limitations and assumptions. Only with these critical changes can the manuscript be adequately evaluated for its scientific merit and clinical relevance.
**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.
Shanghai Xuhui et.al. reported a retrospective study about "Integrating cerebrovascular morphology and radiomics for predicting stroke patients" in a patient population. The work using the ML method is interesting. Overall, the scientific writing meets the requirements of the PeerJ Journal, although there are some typos, e.g., line 130-133, e.g., mm3, and some sentences are either incomplete or missing clarity, e.g., line 86-"Despite....". I suggest replacing the older literature with relatively newer, e.g., 1973 (line 312: studies in media... may not help in scientific discoveries), 2004 (line 306), 2006 (line 299), because I believe the ML method and/or radiomics in MRI were not mature by 1973. Likewise, "sparse representation method" in line 141 needs further explanation for comprehension by all readers. In the feature detection sub-section of the Method section, it is better to refer to your figures for faster reading, e.g., which figures illustrate the 40 morphology of CVS features?
In Figure 1, indicate how many did not participate in the study groups, e.g,. DWI+CTA, 207-153=54?Is this number correct and so on...
Table 2 is labeled Table 1, and Table 1 is labeled Table 2
Explain MRS or mRS in Fig.3.
Put the % values in the confusion matrix, Fig. 4-5
Nomogram and combining make redundant meaning; replace combining with "of". Fig.7
There is no experiment in this study because the data were collected from the retrospective patient population. I believe Figure 2 shows the workflow, framework, or pictorial summary. The objective and aim of this study are cited (Xu et.al.) lines 65-67. If this is from the literature, what are your aims of the study? Citation needs careful attention, differentiating your statement or the literature statement.
This study high impact on the scientific community. Although ML itself is a statistical approach, I suggest including a separate Statistical Analyses section in the Method section. Overall, the writing is good.
Already commented above that citations should be the latest possible. Write clearly what the limitations of this study are. The sentence in line 86," Despite these achievements, limitations remain," seeks what identify the limitations.
-
The manuscript presents significant methodological issues, particularly the absence of a well-defined and valid imaging protocol. The current approach, relying on DWI and ADC MRI alone, is insufficient to accurately differentiate infarct core from penumbra, which is a well-established limitation. A more comprehensive and standardized imaging protocol, including modalities such as CT perfusion, is necessary but currently not adequately implemented or described.
Moreover, the mention of CT perfusion in the text contrasts with the patient selection table, which only reports CT angiography, indicating inconsistencies in the study design.
There are also ethical concerns regarding the imaging workflow. Patients underwent a full CT protocol involving radiation and contrast agent administration, followed by MRI sequences that do not align with accepted acute stroke protocols. This approach may have led to unnecessary delays in initiating urgent treatments such as thrombectomy, which is critical in time-sensitive stroke management.
Overall, the imaging protocol requires substantial improvement and clearer documentation to ensure methodological rigor, reproducibility, and patient safety.
1. As a reviewer, I kindly ask the authors to provide a justification for the selection of these 26 features, specifically, how and why they were chosen to construct the classification model.
2 The phrase 'Many existing methods rely on manual vascular assessments by clinicians' could be clarified by specifying that these assessments are typically performed by radiologists, medical professionals specialized in interpreting radiological images.
1. The term 'cerebral vasculature' could benefit from further clarification. It would be helpful if the authors could specify whether they refer to anatomical structures (e.g., arteries and veins), functional characteristics, or specific vascular features (e.g., branching patterns, vessel diameters) used in their analysis.
2. As a reader and reviewer, I find the Materials and Methods section insufficiently clear. It is not explicitly stated which imaging protocols were performed during the acute phase of stroke — was it non-contrast CT, CT angiography, CT perfusion, or a combination of these? Furthermore, did all patients subsequently undergo MRI? This raises concerns regarding the time efficiency and clinical feasibility of the proposed workflow, especially in an acute setting.
3. Additionally, the criteria for patient inclusion and imaging selection remain vague. It is particularly important to clarify whether only patients with successful reperfusion (TICI 2b–3) were included in the analysis. Without achieving sufficient recanalization, other factors (such as imaging features or vascular characteristics) may have limited clinical relevance. Therefore, a clearer description of the imaging protocol, patient selection criteria, and treatment outcomes is essential to understand the validity and applicability of the study.
It is unclear whether the vascular feature extraction step described (‘we developed a vascular feature extraction pipeline to quantify characteristics such as vascular length, tortuosity, volume, and branching patterns’) was performed manually or fully automatically. Additionally, I would appreciate clarification on whether and how these vascular features were subsequently aligned or integrated with DWI results. Was this registration process automated or manually curated?
4. The current description of the imaging methods used is vague and lacks the necessary detail to ensure reproducibility and clinical interpretability. The statement 'all these cases met the diagnostic criteria for acute cerebral infarction, were diagnosed using imaging such as cranial CT or cranial MRI' does not clearly specify which modality was used for which patients. It is essential to clarify whether non-contrast CT, MRI, or both were performed, and under what circumstances.
Furthermore, while DWI and ADC maps are mentioned, it remains unclear how the infarct core and penumbra were defined. DWI alone does not clearly distinguish between the core and the penumbra. While ADC thresholds may help approximate core areas, this approach is not elaborated upon in the manuscript. In standard clinical and research settings, additional sequences such as perfusion MRI or arterial spin labeling (ASL) are typically necessary to assess penumbra, or, in some cases, a DWI–FLAIR mismatch is used. However, recent studies have criticized the latter approach due to the lack of clear criteria for FLAIR hyperintensity and its temporal dynamics.
As it stands, the Materials and Methods section lacks sufficient detail on the imaging protocol and diagnostic criteria, and must be revised to clearly describe the modalities used, imaging timing, criteria for core/penumbra estimation, and the consistency of these approaches across patients.
5. In the Methods section, you mention the use of CT perfusion (CTP), but it is unclear where this fits into the imaging protocol, as it was not previously described. This raises concerns about the reproducibility of your model, since in standard clinical practice, patients are not routinely imaged with a full CT protocol lasting 7 minutes followed by an MRI. Furthermore, relying solely on DWI and ADC sequences is not an optimal approach for comprehensive evaluation, especially in the acute setting.
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.