Review Reports - An Immersive P300 Brain–Computer Interface Based on 3D Morphological Stimuli and Self-Adaptive Bayesian Linear Discriminant Analysis

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear Authors,

The work done and the presentation are good. Anyhow, a concern needs to be addressed.

1. It is mentioned that 24 participants of the age group 20-25 are considered for the proposed work. Why were 24 participants specifically considered, and especially this age group? Why were the participants aged up to 30 or 35 not considered? What will be the expected difficulties if another age group is considered? Discuss with proper reasoning.

2. How did you handle the noisy issues?

3. Recheck the Section heading 6. Patents.

Author Response

Major Comments：

It is mentioned that 24 participants of the age group 20-25 are considered for the proposed work. Why were 24 participants specifically considered, and especially this age group? Why were the participants aged up to 30 or 35 not considered? What will be the expected difficulties if another age group is considered? Discuss with proper reasoning.

Response: Thank you for this important comment. The inclusion of 24 participants aged 20–25 years was mainly based on the exploratory nature of the present study, experimental feasibility, and the need to control inter-subject variability in an initial VR-P300-BCI validation experiment. The present work aimed to evaluate the feasibility and performance of the proposed 3D-Morph stimulation paradigm and SA-BLDA strategy under relatively controlled conditions. Because each participant completed both EEG recording and immersive VR-based P300 tasks with repeated stimulation trials, the experiment was time-consuming and required stable attention, visual comfort, and tolerance to the VR environment. Therefore, 24 participants were considered appropriate for the initial method-validation stage, although we agree that this sample size is not sufficient to fully establish population-level generalizability. The age range of 20–25 years was selected to reduce age-related variability in visual processing, attentional ability, fatigue resistance, VR adaptation, and ERP characteristics such as P300 amplitude and latency. All participants had normal or corrected-to-normal vision, no color vision deficiency, no self-reported neurological or psychiatric history, and completed VR adaptability screening before the formal experiment. These criteria helped reduce confounding factors unrelated to the proposed stimulation paradigm and classification strategy. Participants aged 30–35 years were not excluded because they are unsuitable; rather, they were not included in this initial phase to maintain a relatively homogeneous cohort. Including broader age groups may introduce several challenges, including increased inter-subject variability in P300 responses, differences in visual comfort and VR tolerance, greater fatigue or cybersickness susceptibility, variability in attention stability and task endurance, and the need for larger sample sizes or age-stratified analyses to maintain sufficient statistical reliability.

How did you handle the noisy issues?

Response: Thank you for this important comment. In this study, noisy issues were handled at multiple stages, including participant screening, EEG acquisition, and offline data processing. First, at the participant-selection stage, all participants had normal or corrected-to-normal vision, no color vision deficiency, and successfully passed the VR adaptability screening before the experiment, which helped reduce noise caused by discomfort, visual mismatch, or unstable adaptation to the VR environment. Second, during EEG acquisition, signals were recorded at a sampling rate of 250 Hz using Ag/AgCl electrodes placed at Fz, Cz, P3, P4, P7, P8, O1, and O2 according to the international 10–20 system. These channels were selected because they cover the frontocentral, parietal, posterior temporal, and occipital regions that are closely related to P300 and visual ERP generation, which improves the signal relevance and helps reduce interference from less informative regions. In addition, to minimize noise contamination during recording, standard EEG quality-control procedures were followed, including careful scalp preparation, impedance control, and instructing participants to reduce excessive blinking, head movement, jaw clenching, and body motion during stimulation. Trials with obvious artifacts or poor signal quality were excluded from further analysis. Furthermore, because the proposed paradigm is based on P300 responses elicited by repeated target stimulation, trial averaging and feature extraction can effectively suppress random noise and enhance ERP-related components. The self-adaptive Bayesian linear discriminant analysis (BLDA) classifier further improves robustness to residual noise by emphasizing discriminative spatiotemporal EEG patterns.

Recheck the Section heading 6. Patents.

Response: Thank you for your careful and helpful comment. We have thoroughly rechecked the content of Section 6 in the revised manuscript. After re-examination, we confirmed that the original heading “Patents” was not appropriate for the present study and did not match the actual content of this section. Accordingly, this issue has been carefully corrected in the revised version. Specifically, we have verified the accuracy of the subsections including Author Contributions, Funding, Informed Consent Statement, and Acknowledgments. Since this study involved human participants and relevant ethical approval has been stated in the manuscript, we have additionally added an independent Institutional Review Board Statement to supplement the ethical compliance information and strictly comply with the journal’s reporting requirements. Furthermore, we revised the Data Availability Statement and Conflicts of Interest sections to ensure that the end matter is accurate, complete, and fully aligned with the actual study conditions. Therefore, the inappropriate “Patents” heading has been corrected, and Section 6 has now been reorganized according to the journal format and the actual content of this manuscript.

We have made the following revisions in the revised manuscript:

“Author Contributions: Conceptualization, J.X. and J.L.; Methodology, J.X. and J.L.; Software, J.L.; validation, J.L.; formal analysis, J.L.; investigation, J.L.; resources, J.X.; data curation, J.L.; writing---original draft preparation, J.L.; writing---review and editing, J.L. and Z.R.; visualization, J.L.; supervision, J.X., Z.R. and X.C.; project administration, J.X. and Z.R.; funding acquisition, J.X. All authors have read and agreed to the published version of the manuscript.

Funding: This research was funded by the Key Project of Hunan Provincial Department of Education under Grant 21A0494, in part by the Natural Science Foundation of Hunan Province, China under Grant 2022JJ50314.

Institutional Review Board Statement: This study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of Guangdong Work Injury Rehabilitation Hospital, China (Protocol No. AF/SC-07/2024.53; approval date: 6 September 2024). All participants were fully informed of the study purpose, procedures, and experimental protocols, and written informed consent was obtained from all subjects prior to participation.

Informed Consent Statement: Informed consent was obtained from all subjects involved in the study.

Data Availability Statement: The data sets presented in this study are available on request from the corresponding author.

Acknowledgments: The authors wish to thank all anonymous reviewers for their constructive comments and suggestions on this manuscript.

Conflicts of Interest: The authors declare no conflicts of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript; or in the decision to publish the results.”(Page 21)

Reviewer 2 Report

Comments and Suggestions for Authors

This paper proposes an immersive P300 brain-computer interface system that combines three-dimensional morphological stimulation (3D-Morph) with self-adaptive Bayesian linear discriminant analysis (SA-BLDA). It has achieved results in improving classification performance, information transfer rate, and reducing user subjective workload. This is an interesting study with innovation and relatively complete research content.

The following are specific review comments and suggestions for revision regarding several issues:

The "comparison benchmark" between offline and online experiments is not entirely equivalent. The offline experiment uses a fixed 10 stimulation rounds to evaluate the performance ceiling and plot the learning curve, while the online experiment uses SA-BLDA to dynamically determine the number of rounds (3-7 rounds). In the discussion section (e.g., Section 4.B), directly comparing the online adaptive results with the peak performance of the 10th offline round cannot isolate the contribution of the adaptive strategy itself, nor does it fairly compare the performance of the two paradigms under "the same decision time". It is recommended to clearly point out the limitations of the current comparison method, explaining that the 10th-round offline result represents performance under "maximum information accumulation", while the online SA-BLDA result is a balanced performance after "efficiency optimization". A more ideal direct comparison would be to also run a set of online experiments with a "fixed number of rounds" (e.g., 5 or 7 rounds). If such data is not available in this study, it should be stated in the text.
The rationale for selecting key parameters in the SA-BLDA algorithm is insufficiently explained. The text mentions that SA-BLDA uses an initial confidence threshold θ₀ = 0.2, but provides no explanation for the basis of selecting this crucial parameter (e.g., based on pilot experiments, grid search, or common values in the field). Arbitrary parameter selection affects the reproducibility of the method and the robustness of the conclusions. It is recommended to briefly supplement the reason for setting this parameter.
Insufficient analysis of individual differences. Table 1 shows considerable performance variations among different subjects. The paper only reports means and standard deviations without any discussion of the potential causes for these differences. This weakens the argument for the system's universality and robustness. It is recommended to add a brief analysis of individual differences. For example, calculate and report correlations between individual performance metrics (e.g., online ACC) and possible influencing factors (e.g., NASA-TLX subscale scores, offline P300 average amplitude). Even if no significant correlation is found, this should also be stated.
Insufficient depth in discussing the interface layout. The paper correctly points out the limitations of the current interface, which follows the traditional 2×4 matrix layout, but fails to further elaborate on the key scientific issues it poses in an immersive 3D environment. In three-dimensional space, the spatial distribution of stimuli (such as depth, non-coplanar arrangement, or surrounding layouts) may fundamentally alter users' visual search strategies and attention allocation patterns, thereby affecting P300 elicitation efficiency and user experience. It is recommended to elevate this from a simple statement of "limitation" to an important direction for future research. This could be an unexplored and highly valuable research direction. Future work could design and compare different three-dimensional spatial layouts to explore optimal design principles for immersive BCI interaction. This discussion would significantly enhance the paper's forward-looking perspective.

Author Response

General Comment: This paper proposes an immersive P300 brain-computer interface system that combines three-dimensional morphological stimulation (3D-Morph) with self-adaptive Bayesian linear discriminant analysis (SA-BLDA). It has achieved results in improving classification performance, information transfer rate, and reducing user subjective workload. This is an interesting study with innovation and relatively complete research content.

Response: Thank you for your careful review of our manuscript and for providing valuable suggestions. We have comprehensively revised the manuscript in accordance with your feedback. We will address each of your comments individually.

The "comparison benchmark" between offline and online experiments is not entirely equivalent. The offline experiment uses a fixed 10 stimulation rounds to evaluate the performance ceiling and plot the learning curve, while the online experiment uses SA-BLDA to dynamically determine the number of rounds (3-7 rounds). In the discussion section (e.g., Section 4.B), directly comparing the online adaptive results with the peak performance of the 10th offline round cannot isolate the contribution of the adaptive strategy itself, nor does it fairly compare the performance of the two paradigms under "the same decision time". It is recommended to clearly point out the limitations of the current comparison method, explaining that the 10th-round offline result represents performance under "maximum information accumulation", while the online SA-BLDA result is a balanced performance after "efficiency optimization". A more ideal direct comparison would be to also run a set of online experiments with a "fixed number of rounds" (e.g., 5 or 7 rounds). If such data is not available in this study, it should be stated in the text.

Response: We sincerely thank the reviewer for this important and constructive comment. We fully agree that the comparison benchmark between the offline and online experiments in the current manuscript is not strictly equivalent. In the offline analysis, we used a fixed 10-round accumulation to estimate the upper-bound performance and to characterize the round-wise learning curve, whereas in the online experiment the SA-BLDA framework dynamically terminated stimulation according to classification confidence, with the decision typically made within 3–7 rounds, as illustrated in the algorithmic workflow in Figure 4. Therefore, the 10th-round offline result reflects performance under maximum information accumulation, while the online SA-BLDA result reflects a trade-off optimized for both accuracy and efficiency, rather than a benchmark under the same decision time. We agree that these two conditions should not be interpreted as directly equivalent. To address this concern, we have revised the manuscript to explicitly state this limitation in the Discussion (Section 4.D) and to avoid overclaiming the adaptive effect based on a direct comparison with the 10-round offline peak.

We have made the following revisions in the revised manuscript:

“Fourth, the offline and online evaluations were not based on strictly equivalent decision rules. The offline experiment used fixed 10-round stimulation to estimate the performance upper bound, whereas the online experiment used SA-BLDA to adaptively stop within 3 to 7 rounds. Therefore, the offline 10th-round result reflects maximum information accumulation, while the online result represents a trade-off between accuracy and interaction efficiency. Future studies should include fixed-round online controls, such as 5- or 7-round conditions, to enable fairer comparisons.” (Page 20, Section E)

The rationale for selecting key parameters in the SA-BLDA algorithm is insufficiently explained. The text mentions that SA-BLDA uses an initial confidence threshold θ₀ = 0.2, but provides no explanation for the basis of selecting this crucial parameter (e.g., based on pilot experiments, grid search, or common values in the field). Arbitrary parameter selection affects the reproducibility of the method and the robustness of the conclusions. It is recommended to briefly supplement the reason for setting this parameter.

Response: We thank the reviewer for this valuable comment. We agree that the rationale for selecting the initial confidence threshold θ₀ should be more clearly described. In the revised manuscript, we have added a detailed explanation in Section 2.6.2 Online Signal Processing. Specifically, we clarified that θ₀ = 0.2 was selected based on pilot experiments, empirical parameter tuning, and the trade-off between classification accuracy and response speed in real-time P300-based BCI control. We further explained that a lower threshold may cause premature incorrect command output, whereas a higher threshold may require more stimulation rounds and reduce the information transfer rate. In addition, we clarified that θ₀ is an initial confidence-control threshold rather than a fixed final decision boundary, and that the final output depends on the accumulated evidence across stimulation rounds, especially the confidence separation between u_max and u_sec. To further improve reproducibility, we also added a parameter table summarizing the definitions and values of the key parameters used in the SA-BLDA algorithm.

We have made the following revisions in the revised manuscript:

“In SA-BLDA, the initial control output threshold was set to θ₀ = 0.2 based on pilot tests on calibration data and preliminary online runs. Several candidate thresholds (such as 0.1, 0.15,0.2, 0.25,0.3) were compared by considering both classification accuracy and the average number of stimulation rounds. Lower thresholds increased the risk of premature incorrect outputs, whereas higher thresholds prolonged stimulation and reduced the information transfer rate. Therefore, θ₀ = 0.2 was selected as a compromise between decision reliability and response speed. Moreover, θ₀ was used as an initial confidence-control threshold rather than a fixed final decision boundary. The definitions, values, and selection rationale of the main SA-BLDA parameters are summarized in Table 1.” (Page 10, Section 2.6.2)

Table I. Key parameters used in the SA-BLDA algorithm.

PARAMETER	DEFINITION	VALUE
θ₀	Initial control output threshold	0.2
∆θ	Confidence-level parameter & Actual value used in the algorithm	Actual value used in the algorithm
L_max	Maximum stimulation rounds	3
L_min U_max U_sec	Minimum stimulation rounds Largest score in regression vector u Second-largest score in regression vector u	7 Computed online Computed online

Insufficient analysis of individual differences. Table 1 shows considerable performance variations among different subjects. The paper only reports means and standard deviations without any discussion of the potential causes for these differences. This weakens the argument for the system's universality and robustness. It is recommended to add a brief analysis of individual differences. For example, calculate and report correlations between individual performance metrics (e.g., online ACC) and possible influencing factors (e.g., NASA-TLX subscale scores, offline P300 average amplitude). Even if no significant correlation is found, this should also be stated.

Response: Thank you for this important suggestion. We agree that the individual differences shown in Table 1 required further analysis and discussion. In the revised manuscript, we have added a correlation analysis between objective online performance and subjective self-rated performance, as shown in Fig. 12.

We have made the following revisions in the revised manuscript:

“In addition, to further examine the individual differences observed in the online experiment, we analyzed the association between objective online performance and subjective self-rated performance (Fig.12). Specifically, the online ACC of each subject was correlated with the NASA-TLX performance subscale score. A strong positive correlation was observed between online ACC and the NASA-TLX performance subscale score across subjects, with a correlation coefficient of r = 0.98. This finding provides evidence that the inter-subject variability in online BCI control was not merely random fluctuation in the decoding results, but was also reflected in the subjects’ subjective experience. Subjects with more stable online recognition performance may have perceived stronger control over the system, resulting in higher self-rated performance.” (Page 19, last paragraph)

Insufficient depth in discussing the interface layout. The paper correctly points out the limitations of the current interface, which follows the traditional 2×4 matrix layout, but fails to further elaborate on the key scientific issues it poses in an immersive 3D environment. In three-dimensional space, the spatial distribution of stimuli (such as depth, non-coplanar arrangement, or surrounding layouts) may fundamentally alter users' visual search strategies and attention allocation patterns, thereby affecting P300 elicitation efficiency and user experience. It is recommended to elevate this from a simple statement of "limitation" to an important direction for future research. This could be an unexplored and highly valuable research direction. Future work could design and compare different three-dimensional spatial layouts to explore optimal design principles for immersive BCI interaction. This discussion would significantly enhance the paper's forward-looking perspective.

Response: Thank you for this insightful and constructive comment. We agree that the interface layout is not merely an implementation issue, but a key scientific factor in immersive P300-based BCI design. In the revised manuscript, we have expanded the “Limitations and Future Work” section to discuss this issue in greater depth.

We have made the following revisions in the revised manuscript:

“Finally, the current interface still followed a conventional 2 $\times$ 4 matrix layout. Although this design preserved comparability with classical P300 interfaces, it did not fully exploit the spatial affordances of immersive VR. Different 3D layouts, such as depth-layered, non-coplanar, circular, or surrounding arrangements, may affect visual search, attention allocation, P300 elicitation, and user experience [50,51]. Future studies should systematically compare these layouts to establish design principles for immersive P300-BCI interfaces.” (Page 20, last paragraph)

[50] Niu, L., Bin, J., Wang, J., Zhan, G., Jia, J., Zhang, L., Gan, Z., and Kang, X. 2023. Effect of 3D paradigm synchronous motion for SSVEP-based hybrid BCI-VR system. Med. Biol. Eng. Comput.61, 9: 2481–2495. DOI: 10.1145/3654777.3676451.

[51] Zhang, L., Pan, J., Gettig, J., Oney, S., and Guo, A. 2024. VRCOPILOT: Authoring 3d layouts with generative ai models in VR. In Proceedings of the 37th Annual ACM Symposium on User Interface Software and Technology, 2024, pp. 1–13.DOI: 10.1007/s11517-023-02845-8

Reviewer 3 Report

Comments and Suggestions for Authors

STRUCTURAL AND ORGANIZATIONAL RECOMMENDATIONS

1.1 Abstract - Needs restructuring

Identified Issues:

- The abstract is too dense and combines methodology, results and conclusions without clear separation

- Key statistical quantifications are missing in the first part

1.2 Introduction - Needs expansion

Current Shortcomings:

- The introduction is relatively short (page 2) compared to the complexity of the topic

- Recent reference studies (2026) on immersive BCI are missing

- There is no conceptual diagram illustrating the gap in the literature

METHODOLOGY - CRITICAL IMPROVEMENTS

2.1 Participants - Insufficient details

Issue: Section 2.1 (Topic) is very brief.

2.2 SA-BLDA Algorithm - Insufficient Technical Details

Missing pseudo-code: The algorithm should be presented as structured pseudo-code:

Computational complexity: The processing time per round for SA-BLDA is not mentioned - critical for real-time applications!

2.3 EEG Signal Processing - Inconsistencies

RESULTS PRESENTATION - MAJOR IMPROVEMENTS

3.1 Offline results - Incomplete statistical analysis

Figure 6 presents the offline results:

Error in interpreting statistics: - The text mentions "significant difference emerges from the 5th round" but the figure shows from round 4

Size effects are missing: - Cohen's d or η² for significant differences

Not presented: - Individual learning curve (inter-subject variability), - Confusion matrix for classifying the 8 commands, - Per-class accuracy (some commands may be more difficult)

DISCUSSION - NEEDS SUBSTANTIAL EXTENSION

4.1 Limitations - too succinct

The limitations mentioned (page 15) are insufficient:

4.2 Future Work - Too generic

TECHNICAL ASPECTS OF WRITING

5.1 Scientific writing - Errors and Improvements

Language Issues Identified: Terminological Inconsistencies:

"3D-Morph paradigm" vs "3D-morphological paradigm";

- "SA-BLDA" vs "self-adaptive Bayesian linear discriminant analysis" – the abbreviation should be defined at first use

Ambiguities:

- "visual fatigue" - operationally defined? Measured how?

- "user engagement" - subjective vs. objective metric?

Voice and tense:

- There are alternations between present and past tense

5.2 Formatting and style

Recommendations:

Figures:

- The authors must increase font size in figures (too small in some cases)

- All figures should be >300 DPI for print

Tables:

- Table 1 is hard to read – consider splitting into 2 tables or landscape format

Equations:

- Equation (1) – authors should define all symbols immediately after

CONCLUSIONS - REFORMULATION

6.1 Current conclusions - Too strong vs. evidence

Problem: The conclusions support "effectiveness confirmed" but:

- Sample size modest (N=24)

- Single-site study

- Short-term evaluation only

- Healthy young participants only

Author Response

Major Comments：

structural and organizational recommendations. (1.1) Abstract - Needs restructuring. The abstract is too dense and combines methodology, results and conclusions without clear separation; Key statistical quantifications are missing in the first part. (1.2) Introduction - Needs expansion. The introduction is relatively short (page 2) compared to the complexity of the topic; Recent reference studies (2026) on immersive BCI are missing; There is no conceptual diagram illustrating the gap in the literature.

Response: Thank you for your valuable suggestions. Each of your comments has been addressed individually, with detailed explanations provided below.

(1.1) We agree that the original abstract was overly condensed and did not clearly separate the study motivation, methodological design, quantitative results, and conclusions. In the paper, we have completely restructured the abstract.

We have made the following revisions in the revised manuscript:

“Conventional P300-based brain-computer interfaces (BCIs) commonly rely on two-dimensional (2D) visual flashing, which may induce visual fatigue and limit immersion, thereby restricting long-term usability and system performance. To address these limitations, this study proposes an immersive P300-BCI framework integrating a three-dimensional morphological stimulation paradigm, termed 3D-Morph, with self-adaptive Bayesian linear discriminant analysis (SA-BLDA). Instead of using color or luminance flickering, the proposed paradigm employs dynamic 2D-to-3D morphological transformations of virtual objects in a virtual reality environment to enhance target-related event-related potentials while preserving visual immersion. SA-BLDA further adjusts the number of stimulation rounds according to classification confidence to balance accuracy and interaction efficiency. Experiments with 24 participants showed that the proposed system outperformed the conventional 2D paradigm. In offline analysis, the proposed method achieved an average classification accuracy of 94.17% and an information transfer rate (ITR) of 25.50 bits/min, significantly outperforming the 2D paradigm (87.29% accuracy, 22.75 bits/min ITR, both p < 0.001, Cohen’s d ≥ 1.22). In online experiments, the 3D-Morph paradigm achieved an average accuracy of 91.46% and an ITR of 37.23 bits/min, compared with 83.96% and 28.74 bits/min for the conventional 2D paradigm (both p < 0.01, Cohen’s d ≥ 1.14). The average response time was reduced by 0.46 s (p < 0.01, Cohen’s d = 0.78), and NASA-TLX evaluations indicated a significantly lower subjective workload across all dimensions (all p < 0.05, Cohen’s d ≥ 0.76). These results demonstrate that combining 3D-Morph stimulation with SA-BLDA can significantly improve classification performance, interaction efficiency, and user experience, providing a feasible framework for immersive and practical P300-BCI applications.”

(1.2) We agree that the original Introduction was relatively concise and did not sufficiently reflect the complexity of immersive P300-based BCI systems. In the revised manuscript, we have substantially expanded the Introduction to provide a more comprehensive and logically progressive background.

We have made the following revisions in the revised manuscript:

“Nevertheless, most existing P300-BCI systems still adopt 2D matrix-based or symbolic visual interfaces. Although these interfaces are effective in laboratory settings, they may not sufficiently represent real-world interaction scenarios, leading to limited immersion, reduced ecological validity, and potential visual fatigue during long-term use.” (Page 2, first paragraph)

“Nevertheless, existing visual stimulation methods mainly rely on color or luminance switching, which may not effectively alleviate visual fatigue. In contrast, using morphological changes of target objects as visual stimuli has been reported to reduce visual fatigue and enhance target-related ERP components. For example, Li et al. proposed a green familiar faces (GFF) spelling paradigm for a P300 speller, in which conventional character flashing was replaced by alternating green familiar faces and characters. Their online results showed that the GFF paradigm elicited larger mean ERP amplitudes than the conventional spelling paradigm, and increased the average classification accuracy from 75.6% to 86.1%. Compared with conventional screen-based interfaces, 3D environments provide embodied and more natural interaction contexts, which may improve user engagement and further enhance ERP responses [34]. In this context, 3D virtual objects can be manipulated through morphological changes, such as scaling, deformation, or structural transformation, thereby providing more natural and spatially meaningful target cues. Such morphology-based stimulation may reduce visual discomfort and enhance attentional engagement [35]. However, most existing immersive BCI studies mainly focus on scene construction, feedback presentation, or task execution, whereas the role of stimulus morphology in eliciting reliable P300 responses remains insufficiently investigated.” (Page 3, first paragraph)

[34] Valente, M., Branco, D., Bermúdez i Badia, S., Fernandes, J.-C., Figueiredo, P., A. 2026. EEG-based predictors of motor recovery during immersive VR-BCI rehabilitation. Sci. Rep. 2026. https://doi.org/10.1038/s41598-026-39106-1 DOI: 10.1038/s41598-026-39106-1.

[35] Arif, S. M. U., Shomanov, A., Tyler, B., and Yazici, A. 2025. Next-generation BCI spellers: a comparative study of ERP responses and fatigue in mixed, virtual, and desktop realities. Virtual Real. 30, 29. https://doi.org/10.1007/s10055-025-01286-y DOI: 10.1007/s10055-025-01286-y.

(1.3) Thank you for this constructive suggestion. We agree that the research gap should be presented more explicitly. In the revised manuscript, instead of adding a separate conceptual diagram, we have strengthened the textual explanation in the Introduction to clearly identify the current limitations in the field and the scientific problem addressed by this study. Specifically, we emphasized that current P300-BCI and immersive BCI studies still have several limitations: conventional P300-BCI systems mainly rely on 2D matrix-based interfaces and color- or luminance-flashing stimuli, which may limit immersion and natural interaction; although 3D virtual scenes have been increasingly introduced into BCI systems, the effect of 3D morphological stimulation on P300 elicitation and classification performance has not been systematically investigated; and many existing systems still adopt fixed stimulation rounds or subject-dependent calibration strategies, making it difficult to simultaneously optimize classification accuracy, response time, and user workload during online interaction. These revisions clarify the literature gap and directly motivate the proposed 3D-Morph paradigm combined with the SA-BLDA strategy, which was designed to improve classification performance, interaction efficiency, and user experience in immersive P300-BCI applications. All corresponding revisions have been highlighted in red in the revised manuscript for the reviewer’s convenience.

Methodology - critical improvements. (2.1) Section 2.1 is very brief. (2.2) SA-BLDA Algorithm - Insufficient Technical Details. Missing pseudo-code: The algorithm should be presented as structured pseudo-code; Computational complexity: The processing time per round for SA-BLDA is not mentioned - critical for real-time applications. (2.3) EEG Signal Processing – Inconsistencies.

Response: Thank you for your valuable and constructive comments regarding the methodological clarity and technical completeness of our manuscript. We fully agree that a rigorous description of the participant information, SA-BLDA implementation, computational feasibility, and EEG signal-processing pipeline is essential for improving the reproducibility and scientific quality of this study. In response to these comments, we have carefully revised the Methodology section of the manuscript and made the following major improvements: (2.1) expanded the description of participants and ethical procedures; (2.2) added structured pseudo-code and implementation details for the SA-BLDA algorithm; and (2.3) clarified and unified the EEG signal acquisition and processing pipeline. These revisions are intended to make the proposed immersive P300-BCI framework more transparent, reproducible, and suitable for real-time application scenarios.

(2.1) Thank you for this helpful suggestion. We agree that the original description of the participant information was too brief. In the revised manuscript, we have expanded Section 2.1 to provide a clearer description of the participants and the inclusion criteria. Specifically, we added information regarding handedness, visual and auditory conditions, color vision, psychiatric history, severe visual disorders, medication status, and the pre-experiment VR adaptability screening and task familiarization procedure. The revised content has been highlighted in red in the manuscript.

We have made the following revisions in the revised manuscript:

“A total of 24 healthy, right-handed participants took part in this study, including 16 males and 8 females, aged 20–25 years. All participants had normal or corrected-to-normal vision and normal hearing, no color vision deficiency, and no self-reported history of neurological, psychiatric, or severe visual disorders. None of them reported taking medication that could affect cognitive function or EEG responses. Before the formal experiment, all participants completed a brief VR adaptability screening and task familiarization session to ensure that they could tolerate the immersive environment and fully understand the experimental procedure.

This study was approved by the Ethics Committee of Guangdong Work Injury Rehabilitation Hospital, China (protocol number: AF/SC-07/2024.53; approval date: September 6, 2024)., and all participants provided written informed consent before participation. All procedures were conducted in accordance with the Declaration of Helsinki.” (Page 4, Section 2.1, first paragraph)

(2.2) Thank you for this constructive suggestion. We agree with the reviewer that the original manuscript did not provide sufficient implementation details for the SA-BLDA algorithm. Since SA-BLDA is a key component of the proposed immersive P300-BCI system, presenting the algorithm as structured pseudo-code is important for improving the clarity, reproducibility, and technical completeness of the Methodology section. In response to this comment, we have revised the SA-BLDA subsection by adding a structured pseudo-code flow, entitled “Algorithm 1. Procedure of the SA-BLDA Algorithm.” The added pseudo-code explicitly describes the complete adaptive decision procedure, including the initialization of stimulation rounds, feature vector and feature matrix updating, BLDA-based regression score calculation, identification of the maximum and second-largest regression scores, confidence-difference calculation, adaptive stopping judgment, and final target output. This directly clarifies how SA-BLDA dynamically determines whether to continue collecting additional stimulation rounds or output the current target. In addition, to avoid ambiguity in the pseudo-code and Figure 4, we have added a parameter-definition table in the revised manuscript. This table summarizes the main variables used in Algorithm 1 and Figure 4. By providing both the structured pseudo-code and the corresponding parameter definitions, the revised manuscript now presents the SA-BLDA implementation in a clearer and more reproducible form.

We also revised the explanatory text in the SA-BLDA subsection to explicitly link the conceptual workflow and the pseudo-code. Specifically, the revised manuscript now states that the structured pseudo-code of the detailed implementation is provided in Algorithm 1, while the conceptual workflow of SA-BLDA is illustrated in Figure 4.

We have made the following revisions in the revised manuscript:

“To improve real-time interaction efficiency while maintaining classification accuracy, a self-adaptive Bayesian linear discriminant analysis (SA-BLDA) algorithm was adopted. Unlike conventional fixed-round P300 classification strategies, SA-BLDA dynamically adjusts the number of stimulation rounds required for decision making based on classification confidence, thereby achieving a balance between accuracy and response speed. The structured pseudo-code of the detailed implementation is provided in Algorithm 1, and the conceptual workflow of SA-BLDA is illustrated in Figure 4. The definitions of the main parameters used in Algorithm 1 and Figure 4 are summarized in Table 1. By adaptively terminating stimulation once sufficient confidence is achieved, the SA-BLDA algorithm effectively reduces unnecessary stimulation rounds, shortens response time, and enhances the ITR, making it particularly suitable for real-time P300-based BCI applications.” (Page 7, Section 2.5.2, first paragraph)

On the order hand, we also agree that the processing time per stimulation round is critical for evaluating the real-time feasibility of SA-BLDA in online P300-BCI applications. In the revised manuscript, we have added the processing time per stimulation round (PT) as an additional evaluation metric in the online experimental results. Specifically, PT was defined as the computational time required by SA-BLDA for one stimulation round, measured from the moment when the EEG data/features of the current round were input into the algorithm to the moment when the classification output and confidence-based stopping decision were generated. For each participant, PT was calculated as the average processing time across all stimulation rounds completed during the 20 online experimental tasks. Therefore, the reported PT reflects the participant-level average computational latency of SA-BLDA during online operation, and the final values were summarized across all 24 participants.

We have made the following revisions in the revised manuscript:

“Moreover, the PT of SA-BLDA was also significantly reduced under the 3D-Morph paradigm, decreasing from 48.54 ± 10.47 ms in the 2D paradigm to 26.40 ± 9.41 ms in the 3D-Morph paradigm (p < 0.01, Cohen’s d = 2.34), corresponding to a 45.61% reduction in computational time per round.” (Page 18, first paragraph)

“The effectiveness of the adaptive stopping strategy was further validated by the online results. As reported in Table 1, under the 3D-Morph paradigm, the SA-BLDA algorithm determined an average of 4.76 ± 0.37 stimulus rounds. Under comparable conditions, the online ACC, 92.50%, was substantially higher than that of the offline result at the fifth round, 82.08%. Although the online ACC, approximately five rounds, was slightly lower than the best offline performance achieved at the tenth round, 94.17%, the ITR was significantly improved, 37.23 bits/min vs. 25.50 bits/min. More importantly, SA-BLDA achieved this adaptive decision process with a low computational cost. The PT under the 3D-Morph paradigm was only 26.40 ± 9.41 ms per stimulation round, indicating that the confidence-based stopping decision could be completed rapidly during online operation without introducing substantial additional latency.

A similar trend was observed for the 2D paradigm. As reported in Table 2, the average number of adaptive rounds in the online experiment was 5.30 ± 0.64, with an ACC, 84.58%, higher than that of the offline sixth-round condition, 75.83%. Although slightly lower than the best offline ACC at the tenth round, 87.29%, the online experiment achieved a higher ITR, 28.74 bits/min vs. 22.75 bits/min. On the other hand, the PT of SA-BLDA in the 2D paradigm was 48.54 ± 10.47 ms per stimulation round, which was significantly higher than that in the 3D-Morph paradigm, 26.40 ± 9.41 ms, p < 0.01, Cohen’s d = 2.34. This corresponds to a 45.61% reduction in per-round processing time under the 3D-Morph paradigm. Considering both the reduced number of stimulation rounds and the shorter PT, the total computational cost required for online command recognition was further decreased, which is beneficial for maintaining system responsiveness in real-time BCI applications.” (Page 17, last paragraph to Page 18, first paragraph)

“Experimental results showed that, at the group level, the proposed 3D-Morph paradigm achieved higher average ACC and ITR than the conventional 2D paradigm in both offline and online evaluations. In the offline experiments, the 3D-Morph paradigm achieved an average ACC of 94.17% and an ITR of 25.50 bits/min, compared with 87.29% and 22.75 bits/min for the 2D paradigm. In the online experiments, the 3D-Morph paradigm achieved an average ACC of 91.46% and an ITR of 37.23 bits/min, corresponding to improvements of 7.50% and 29.54%, respectively. Moreover, the average RT was reduced from 6.33 s to 5.87 s, suggesting improved response speed under the current online task setting. In addition, the processing time per stimulation round for the SA-BLDA algorithm was substantially reduced from 48.54 ms in the 2D paradigm to 26.40 ms in the 3D-Morph paradigm, representing a 45.61% reduction. This result indicates that the proposed paradigm not only improved recognition performance but also reduced the computational time required for online decision making, thereby supporting more efficient real-time BCI operation. Furthermore, subjective workload evaluation using NASA-TLX revealed that the 3D-Morph paradigm was associated with lower subjective temporal demand, perceived effort, and frustration, as well as better self-rated performance, compared with the 2D paradigm. These findings suggest that the proposed approach may improve selected BCI performance metrics and reduce perceived workload under the present experimental conditions” (Page 20, Section 5, second paragraph)

(2.3) Thank you for pointing this out. We have revised the EEG signal processing description to remove the inconsistency between the offline and online procedures. In the revised Section 2.3, we clarified that the EEG processing pipeline consisted of two connected stages: offline calibration and online recognition. Importantly, both stages used the same preprocessing and feature extraction procedures, including baseline correction, band-pass filtering, downsampling, and feature normalization. The offline stage was used to train an individualized BLDA model, whereas the online stage applied the identical processing pipeline and used the offline-trained BLDA model to obtain regression scores for real-time command recognition. These scores were then used by the SA-BLDA algorithm to adaptively determine the stimulation rounds and output the target command. This revision makes the relationship between offline training and online recognition clearer and ensures methodological consistency.

We have made the following revisions in the revised manuscript:

The EEG signal processing pipeline consisted of two connected stages: offline calibration and online recognition. To ensure consistency between the two stages, the same preprocessing and feature extraction procedures were applied, including baseline correction, band-pass filtering, downsampling, and feature normalization. During offline calibration, the extracted features were used to train an individualized BLDA model, which mapped EEG feature vectors to regression scores. During online recognition, real-time EEG epochs were processed using the identical pipeline and then fed into the offline-trained BLDA model. The resulting regression scores were further used by the SA-BLDA algorithm to adaptively determine the number of stimulation rounds and generate the target command in real time. (Page 9, Section 2.6)

RESULTS PRESENTATION - MAJOR IMPROVEMENTS. (3.1) Error in interpreting statistics: The text mentions "significant difference emerges from the 5th round" but the figure shows from round 4. (3.2) Size effects are missing: Cohen's d or η² for significant differences. (3.3) Not presented: Individual learning curve (inter-subject variability), Confusion matrix for classifying the 8 commands, Per-class accuracy (some commands may be more difficult)

Response: We thank the reviewer for the constructive comments regarding the presentation of the offline results.

(3.1) We agree that the original Section 3.1 did not provide a sufficiently complete statistical analysis and contained an inconsistency between the textual description and Figure 6. After rechecking the statistical annotations, we corrected the statement that the significant difference emerged from the 5th round; the revised manuscript now states that the significant difference was first observed from the 4th stimulation round, as shown in Figure 6.

(3.2) We appreciate the reviewer’s valuable suggestion. We fully agree that reporting effect sizes is essential for interpreting the practical magnitude of statistically significant differences, rather than relying solely on p-values. In the revised manuscript, we have added Cohen’s d values for the significant comparisons between the 2D and 3D-Morph paradigms. Additionally, we have revised Section 2.7.3 (Statistical Analyses) to explicitly describe this analytical procedure.

We have made the following revisions in the revised manuscript:

“Given that all 24 participants completed both the 2D and 3D-Morph paradigm experiments, the data constituted paired observations. Accordingly, a two-tailed paired-samples t-test was employed to assess statistical significance for all performance metrics and task workload measures. Additionally, Cohen’s d was reported to assess the practical significance of the observed differences between conditions.” (Page 11, Section 2.7.3)

(3.3) Thank you for this constructive comment. We agree that the original manuscript did not sufficiently present the subject-level performance variation. To address this issue, we have added a round-wise offline accuracy analysis across all 24 subjects in the revised manuscript, as shown in Fig. 11.

In Fig. 11, each scatter point represents the ACC of an individual subject in a given stimulation round, the boxplot summarizes the inter-subject distribution, and the red line indicates the mean ACC across all subjects. The results show that the offline ACC increased progressively with the number of stimulation rounds. In the early rounds, the ACC values were more widely distributed across subjects, indicating clear inter-subject variability. As the number of rounds increased, the distribution became more concentrated and the average ACC improved, suggesting that additional stimulus accumulation reduced individual differences to some extent, although it did not completely eliminate them.

We also examined the online recognition performance of the eight button commands across the 24 subjects. The stimulation interface consisted of eight functional buttons arranged in a 2 × 4 matrix, corresponding to Turn Left, Forward, Backward, Turn Right, Speed Up, Reset, Stop, and Slow Down [1]. It should be noted that each subject completed 20 online control tasks, and the actual command sequence depended on the subject’s online control process. Therefore, the occurrence frequency of the eight button commands was not strictly balanced across subjects. Under this condition, command-wise accuracy should be interpreted together with the number of command occurrences, since commands with fewer occurrences may show more variable accuracy estimates.

As summarized in the table below, the occurrence frequencies of the eight commands were different in the online experiment. Specifically, forward appeared most frequently, with 80 occurrences, whereas Backward appeared least frequently, with only 46 occurrences. The corresponding average ACCs were 83.71% for Forward and 71.25% for Backward. Because the Backward command occurred less frequently, its accuracy estimate was more susceptible to variability caused by the limited number of trials, which may partly explain its relatively lower average ACC. In contrast, the average ACCs of the other commands were generally comparable, including Reset (58 occurrences, 81.25%), Stop (62 occurrences, 91.88%), Decelerate (53 occurrences, 88.88%), Accelerate (70 occurrences, 77.79%), Turn Right (51 occurrences, 88.21%), and Turn Left (60 occurrences, 82.83%).

To further visualize the command results, we generated an additional heatmap in this response letter. The columns represent subjects S1–S24, and the rows represent the eight commands. Each cell shows the online recognition accuracy of the corresponding command for a given subject, and the last column reports the mean recognition accuracy of each command across all 24 subjects. The heatmap shows that, except for the frequency-related fluctuation observed for Backward, the recognition accuracies of the remaining commands were generally close, and no significant difference was observed among the eight commands. Therefore, the current online results do not indicate that any particular command was systematically more difficult to recognize.

Command	Frequency	Average ACC
Reset	58	81.25%
Stop	62	91.88%
Decelerate	53	88.88%
Accelerate	70	77.79%
Turn right	51	88.21%
Turn left	60	82.83%
Forward	80	83.71%
Backward	46	71.25%

We have made the following revisions in the revised manuscript:

“We analyzed the round-wise offline accuracy across all subjects to visualize the subject-level performance variation as the number of stimulation rounds increased. As shown in Fig.11, the round-wise offline ACC increased progressively with the number of stimulation rounds. Each scatter point represents the ACC of an individual subject in a given round and the red line indicates the mean ACC across all 24 subjects.

However, the scatter distribution and boxplots also revealed clear inter-subject variability. In the early rounds, the ACC values were widely distributed across subjects, indicating that some subjects could achieve relatively stable recognition with fewer stimulation rounds, whereas others required more repetitions. As the number of rounds increased, the distribution became more concentrated and the average ACC improved, suggesting that additional stimulus accumulation reduced, but did not completely eliminate, individual differences.” (Page 18 - Page 19, Section D, first paragraph)

Results presentation - major improvements. Limitations is too succinct and future work is too generic

Response: We thank the reviewer for pointing out that the limitations and future work in the original Discussion section were too succinct and generic. We agree with this comment. In the revised manuscript, we have substantially expanded the subsection “D. Limitations and Future Work” on page 15. The revised version now discusses the limitations of the present study from multiple perspectives, including algorithmic parameter selection, experimental comparability, online control conditions, VR interaction layout, long-term usability, and real-world validation.

In addition, we expanded the discussion of VR interface design. Although the proposed 3D-Morph paradigm used realistic 3D objects, the spatial arrangement of stimulation targets still largely followed a conventional matrix-based layout. We now discuss that immersive VR allows alternative spatial layouts, such as depth-layered, arc-shaped, circular, and surrounding arrangements, which may affect visual search, attention allocation, P300 elicitation, classification performance, and user comfort. We further added that future studies should evaluate long-term usability, including visual fatigue, attentional fluctuation, cybersickness, and cross-session robustness.

Finally, we clarified that the current evaluation was limited to healthy young participants in a virtual environment. Future work will integrate the proposed system with physical assistive devices, such as robotic arms, powered wheelchairs, or smart-home control systems, and validate the system with users from different age groups and clinical populations. These revisions make the limitations more comprehensive and the future work more concrete, testable, and directly connected to the findings of the present study.

Technical aspects of writing. (5.1) Scientific writing - Errors and Improvements. Language Issues Identified: “3D-Morph paradigm” vs “3D-morphological paradigm”; “SA-BLDA” vs “self-adaptive Bayesian linear discriminant analysis” – the abbreviation should be defined at first use. Ambiguities: "visual fatigue" - operationally defined? Measured how? “User engagement” - subjective vs. objective metric? Voice and tense: There are alternations between present and past tense. (5.2) Terminological Inconsistencies: Formatting and style. Figures: The authors must increase font size in figures (too small in some cases), all figures should be >300 DPI for print. Tables: Table 1 is hard to read – consider splitting into 2 tables or landscape format. Equation (1) – authors should define all symbols immediately after.

Response: Thank you for your valuable suggestions. Each of your comments has been addressed individually, with detailed explanations provided below.

(5.1) To address the reviewer’s concerns regarding scientific writing, we revised the manuscript from three aspects: terminology consistency, clarification of ambiguous expressions, and tense consistency.

First, we unified the terminology throughout the manuscript. The proposed paradigm is now consistently referred to as the “3D-Morph paradigm,” and inconsistent expressions such as “3D-morphological paradigm” have been replaced. The abbreviation “SA-BLDA” is now defined at first use as “self-adaptive Bayesian linear discriminant analysis (SA-BLDA)” and is used consistently thereafter.

Second, we would like to clarify that visual fatigue and user engagement were not operationally defined as independent outcome variables in the present study, nor were they quantitatively measured using subjective questionnaires, behavioral indices, or physiological indicators. In this study, these terms were discussed only as potential usability-related factors associated with immersive VR-based P300-BCI interaction, rather than as directly measured experimental outcomes. Therefore, we did not conduct statistical analyses or draw quantitative conclusions regarding visual fatigue or user engagement.

Third, we revised the manuscript to improve tense consistency. Experimental procedures and results are now mainly described in the past tense, while general scientific statements and interpretations are described in the present tense where appropriate.

(5.2) To address the reviewer’s comments regarding formatting and style, we revised the figures, tables, and equation descriptions to improve readability and publication quality.

First, we improved the formatting and readability of the figures and tables. The font sizes of figure labels, axes, legends, and annotations were increased, and all figures were regenerated or exported at a resolution of at least 300 DPI.

Second, in response to the reviewer’s suggestion, the original Table 1 was split into two separate tables to reduce information density and improve readability. Accordingly, all in-text references and descriptions related to the original Table 1 were revised to match the new table numbering and content.

Third, we added definitions of all symbols immediately after Equation (1). This revision ensures that readers can understand the mathematical formulation without searching for symbol definitions elsewhere in the manuscript. Overall, these changes improve the clarity, consistency, and readability of the manuscript.

Conclusions - reformulation. (6.1) Current conclusions - Too strong vs. evidence. (6.2) Problem: Sample size modest (N=24); Single-site study; Short-term evaluation only; Healthy young participants only

Response: Thank you for your valuable suggestions. Each of your comments has been addressed individually, with detailed explanations provided below.

(6.1) We thank the reviewer for this important comment. We agree that some statements in the original conclusion were stronger than what can be directly supported by the present experimental evidence. We have revised the Conclusion section to make the claims more cautious and evidence-based.

We have made the following revisions in the revised manuscript:

“In this study, an immersive P300-based BCI system integrating the 3D-Morph stimulation paradigm and the SA-BLDA adaptive classification strategy was proposed. By replacing conventional color-flashing stimuli with 2D-to-3D morphological transformations, the proposed paradigm was designed to provide a more visually dynamic stimulation form and to elicit target-related ERP responses under the tested P300-BCI setting. In addition, the SA-BLDA strategy adaptively adjusts the number of stimulus rounds based on classification confidence, which was associated with favorable ACC and ITR values in the present experiments.

Experimental results showed that, at the group level, the proposed 3D-Morph paradigm achieved higher average ACC and ITR than the conventional 2D paradigm in both offline and online evaluations. In the offline experiments, the 3D-Morph paradigm achieved an average ACC of 94.17% and an ITR of 25.50 bits/min, compared with 87.29% and 22.75 bits/min for the 2D paradigm. In the online experiments, the 3D-Morph paradigm achieved an average ACC of 91.46% and an ITR of 37.23 bits/min, corresponding to improvements of 7.50% and 29.54%, respectively. Moreover, the average RT was reduced from 6.33 s to 5.87 s, suggesting improved response speed under the current online task setting. In addition, the processing time per stimulation round for the SA-BLDA algorithm was substantially reduced from 48.54 ms in the 2D paradigm to 26.40 ms in the 3D-Morph paradigm, representing a 45.61% reduction. This result indicates that the proposed paradigm not only improved recognition performance but also reduced the computational time required for online decision making, thereby supporting more efficient real-time BCI operation. Furthermore, subjective workload evaluation using NASA-TLX revealed that the 3D-Morph paradigm was associated with lower subjective temporal demand, perceived effort, and frustration, as well as better self-rated performance, compared with the 2D paradigm. These findings suggest that the proposed approach may improve selected BCI performance metrics and reduce perceived workload under the present experimental conditions.

Overall, the integration of the 3D-Morph stimulation paradigm and the SA-BLDA adaptive strategy represents a feasible approach for improving selected efficiency-related metrics of P300-based BCI systems, while potentially reducing subjective workload. However, the generalizability of these findings remains to be further validated. Future studies with larger cohorts, more diverse user groups, longitudinal evaluations, and application-oriented tasks are needed to determine the robustness and practical utility of the proposed approach.” (Page 21, Section 5)

(6.2) Thank you for highlighting these important limitations. We have revised the “Limitations and Future Work” to more explicitly address the sample size, participant characteristics, single-site design, and short-term evaluation, and to provide more specific future research directions.

First, we have acknowledged the limitations related to the modest sample size and the inclusion of only healthy young adults, and we now state that future studies should recruit larger and more diverse cohorts, including clinical users.

We have made the following revisions in the revised manuscript:

“First, the present study was limited to 24 healthy young participants aged 20--25 years, restricting the generalizability of the findings. Future work should recruit larger and more diverse populations, including clinical users with motor impairments, and validate the system in real-world assistive-control tasks using practical outcomes such as task completion time, error recovery, workload, safety, and user acceptance.” (Page 20, second paragraph)

Second, we have added the single-site design as a limitation and proposed future multi-center validation under standardized experimental protocols.

We have made the following revisions in the revised manuscript:

“Second, the experiment was conducted at a single site with a fixed VR-BCI hardware configuration. Moreover, visual fatigue, engagement, comfort, cybersickness, and workload were not quantified using standardized questionnaires or objective indicators. Future studies should combine subjective scales with eye-tracking, blink rate, fixation patterns, pupil diameter, behavioral responses, and physiological measures to assess usability more comprehensively.” (Page 20, second paragraph)

Third, we have clarified that the current evaluation was short-term only and proposed longitudinal and multi-session studies to assess long-term stability, learning effects, and fatigue accumulation.

We have made the following revisions in the revised manuscript:

“Third, the present study only evaluated short-term performance. Long-term stability, learning effects, user adaptation, and fatigue accumulation across repeated sessions remain unclear. Future multi-session studies are needed to determine whether the proposed 3D-Morph paradigm and SA-BLDA strategy can maintain stable performance over time.” (Page 20, third paragraph)

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have confirmed the questions I raised and made necessary revisions to the paper.

Reviewer 3 Report

Comments and Suggestions for Authors

The paper can be published in the present form