Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

 

Very insightful and detailed introduction. However, it would be beneficial for the authors to elaborate more on the motivation and importance of their work, similarly to the abstract.

 

The authors have used a very-well known technique (PRISMA) and have also described in depth how the final choice of papers was conducted with relevant details and diagrams.

 

The authors did a very good job describing the most relevant studies for their paper, also summarising them in a very useful to the reader, table.

 

It would have been beneficial for the authors to also include a section with regards to background knowledge on Machine/Deep learning and how it is applied on EEG Microstates

 

In their discussion section, the authors did a very good job explaining and assessing the work conducted in the related studies and effectively comparing them with each other. However, it would have been beneficial to also discuss the related studies from a machine learning pov and ultimately provide an assessment of the good/bad points on the techniques chosen and what improvements could be made. Ultimately, the improvements could be linked with the future directions section. Concluding, the authors should expand their discussion section by also considering the technical part of the paper, machine/deep learning.

Author Response

Thank you very much for the valuable suggestions and insights provided. We are attaching the point-by-point response to your comments. 

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

This systematic review aims to gather current knowledge on machine learning applied to EEG microstate analysis in psychotic disorders. The results show that EEG microstates can be used to accurately classify diagnoses within the psychosis spectrum, across all stages, outperforming models based on conventional EEG measures, with a prominent role of microstate D. One study also suggests that microstate anomalies may be directly linked to symptoms severity. Integrating EEG microstates with machine learning shows promise in improving our understanding of psychotic disorders and developing more precise diagnostic tools. The evidence supports the hypothesis that microstate analysis can discriminate between patients and controls, even in the early stages of the disease and in individuals at high risk.

While the article is well-written and the data is solid, several aspects require further clarification and refinement to enhance the overall clarity and comprehensiveness:

1. Clear Prediction: The authors should include a posible prediction based on the current state of the art. Specifically, the study should outline expectations regarding the outcomes, particularly in terms of differentiating between microstates in psychotic disorders using EEG Microstates and Machine Learning.
2. Clarification of Terms:
• FESZ: When the term FESZ is introduced (line 188), the definition is not provided, although it is mentioned later in the text (line 223). A brief explanation of this term should be given upon its first mention.
• UHR: Similarly, the term UHR is introduced in line 188 without definition, but the meaning is provided later (line 223). A definition should be included when it first appears.
• FES (line 196), FEP (line 203), and ERP (line 233): These terms should also be clearly defined upon their first mention. For example, ERP could be expanded as Event-Related Potential if that is the correct interpretation.
3. Clarification of Microstates:
• Microstate 7, 8, etc.: The terms Microstate 7 (line 235) and others like it should be clearly identified in terms of their canonical category. For example, it should be specified whether Microstate 7 falls under categories A, B, C, or D.
• The term N100-P300 should be defined to clarify whether it refers to specific positive or negative waves of the ERP.
4. Methodological Clarifications:
• MMN-P3a Tasks: The authors should define what the MMN-P3a tasks consist of.
• BPRS Scores: The definition and use of BPRS scores should be clarified to ensure readers understand their significance in the context of the study.
• Z and D Time Series: More specific information on the Z and D time series should be provided to ensure the reader fully understands their role in the analysis.
5. AUC and Support Vector Machine (SVM):
• AUC and AUCs: The term AUC and its variant AUCs (lines 204 and 195) should be clearly defined when first introduced. These terms should also be explained in more detail, as they are critical to the analysis and interpretation of results.
• SVM: The abbreviation SVM (Support Vector Machine) is introduced at line 168 and appears repeatedly in the manuscript (lines 182, 211, 227). It is essential that the full term is defined at the beginning, with the acronym SVM used only in subsequent references.
6. Potential Confounding Factors:

The authors might want to discuss whether external factors, such as sex (e.g., the luteal phase in females) and hand dominance, could influence the results. These factors may have an impact on EEG microstate analysis and should be considered in the interpretation of findings.

 

Author Response

Thank you very much for the valuable suggestions and insights provided. We are attaching the point-by-point response to your comments. 

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

The paper contributes to the field by synthesizing findings from various studies and highlighting the potential of EEG microstates as diagnostic markers. However, the review could enhance its contribution by explicitly addressing gaps in existing research and proposing actionable recommendations for future studies.

The manuscript is well-structured and follows a logical progression. However, the exclusion criteria are generally described but lack specific examples of what exactly was meant by "the use of machine learning solely for preprocessing or processing EEG data." No information is provided regarding which specific types of studies or methods (e.g., particular algorithms) were excluded based on abstracts and titles.

The use of author initials (e.g., F.P., G.G.) in Section 2.2 is unconventional and detracts from the objectivity of the methodology. It is recommended to replace these references with neutral descriptions (e.g., 'two independent reviewers'). Information about individual contributions is included in the 'Author Contributions' section.

 

The tables and figures effectively summarize key findings, enhancing comprehension. However, the column labeled "ML features" in Table 2 is misleading, as it includes more than just features. It also encompasses machine learning algorithms, validation strategies, and data processing techniques, such as dimensionality reduction and semantic modeling. A more accurate label, such as "ML Methods and Features" would better reflect the diverse information presented. The column 'Results' in Table 2 is generally appropriate, as it provides key performance metrics such as accuracy and AUC. However, consistency in the level of detail across all entries would further improve the table's utility. The addition of a new column titled 'Interpretation and Research Implications' could enhance Table 2 by providing context for the numerical results. This column could include the significance of findings, limitations of the study, and suggestions for future research directions. 

The review is well-cited, with references to teh studies in the field. However, it could include a more explicit comparison with prior meta-analyses or systematic reviews.

Author Response

Thank you very much for the valuable suggestions and insights provided. We are attaching the point-by-point response to your comments. 

Author Response File: Author Response.docx