AI-Enabled Secure and Scalable Distributed Web Architecture for Medical Informatics

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Summary of the Study
The manuscript presents a novel distributed web architecture for medical informatics, integrating artificial intelligence (AI), containerized microservices (via Docker Swarm), and HL7 FHIR interoperability standards. The proposed architecture is validated through a simulated deployment involving diffusion tensor imaging (DTI) data, with performance metrics (e.g., F1 score, latency) and CRM-enhanced patient engagement tools. The study aims to address scalability, security, and real-time decision-making in multi-institutional eHealth settings.

Major Concerns and Suggestions
The use of simulated DTI and EHR data is understandable, but the manuscript should clearly explain the limitations of using synthetic data (e.g., lack of noise, variability) and how this may affect generalizability.
It is not clearly stated whether the annotated DTI datasets are public, real, or synthetic. This is crucial for reproducibility and external validation.
The metrics focus on CNN classification only. The system’s overall robustness (e.g., fault tolerance under failure, network latency across nodes) needs further quantitative exploration.
The CNN architecture is briefly described. Including a schematic diagram and training/validation split ratios would enhance clarity.
While Table 1 compares architectures, it lacks detailed methodology. Were these numbers from the literature, or were they experimentally tested on the same setup?
The study would benefit from releasing partial code or simulated datasets to support reproducibility.
The manuscript mentions a "blockchain-inspired ledger" without specifying the implementation framework, consensus model, or scalability impact.
Although TLS 1.3 and encryption are mentioned, a more detailed discussion of GDPR/HIPAA compliance or role-based access control (RBAC) models would strengthen the privacy assurance.
While CRM modules are mentioned frequently, a real-world use case (e.g., patient follow-up reminder, telemedicine integration) would demonstrate the actual clinical utility.
Statements about user satisfaction are anecdotal. Even small-scale usability tests with real users (e.g., SUS score) would add rigor.
The text inside the “CCN Encoder” box is hard to read. Consider enlarging the font or changing the color contrast.
If this refers to the left-panel CRM view, it contains empty or underutilized space. Consider adding more clinical dashboard widgets or workflow illustrations.
Line 36: "Sfat et. all" → "Sfat et al."
Line 109: "dimensionality reduction in DTI (dif- fusion tensor imaging)" → fix hyphenation.
Add 95% confidence intervals for precision, recall, F1-score, and accuracy.
Consider including p-values or tests (e.g., paired t-test or Wilcoxon) comparing your method with baseline architectures.
The manuscript would benefit from a calibration curve or Brier score to assess the reliability of model probabilities.
Define all acronyms at first mention (e.g., PCA, CRM, HIS).
Ensure references [42]–[46] align with content (e.g., LLMs, edge AI). If not, briefly clarify how these support future directions.

Author Response

Suggestion 1:

The use of simulated DTI and EHR data is understandable, but the manuscript should clearly explain the limitations of using synthetic data (e.g., lack of noise, variability) and how this may affect generalizability.

Response 1:

It should be noted that the use of synthetic DTI data presents certain limitations including reduced noise variability, standardized patterns, and potentially less complex anomaly presentations compared to real clinical data. These factors may affect the generalizability of our results to real-world clinical environments where data exhibits greater heterogeneity and unpredictable variations. Future validation with real clinical datasets will be essential to confirm the robustness of the proposed architecture.

 

Suggestion 2:

It is not clearly stated whether the annotated DTI datasets are public, real, or synthetic. This is crucial for reproducibility and external validation.

Response 2:

The model was trained on synthetic annotated DTI datasets generated specifically for this study.

 

Suggestion 3:

The metrics focus on CNN classification only. The system’s overall robustness (e.g., fault tolerance under failure, network latency across nodes) needs further quantitative exploration.

Response 3:

A new subsection titled “System Robustness Evaluation” was added. It includes quantitative results on fault tolerance (99.2% uptime), network latency (10ms–500ms tests), load distribution (100 concurrent requests), and recovery time (MTTR: 3.4s).

 

Suggestion 4:

The CNN architecture is briefly described. Including a schematic diagram and training/validation split ratios would enhance clarity.

Response 4:

The architecture description was expanded to include all layers, filter sizes, activation functions, and training parameters (80/10/10 split, Adam optimizer, batch size 32, early stopping).

 

Suggestion 5:

While Table 1 compares architectures, it lacks detailed methodology. Were these numbers from the literature, or were they experimentally tested on the same setup?

Response 5:

All comparison metrics in Table 1 were obtained through experimental testing on a 4-node cluster using the same synthetic dataset, ensuring fairness.

 

Suggestion 6:

The manuscript mentions a “blockchain-inspired ledger” without specifying the implementation framework, consensus model, or scalability impact.

Response 6:

We specified that a permissioned blockchain using Hyperledger Fabric with PBFT consensus and smart contracts is used. This introduces ~15–20ms latency per transaction and requires additional resources.

 

Suggestion 7:

Although TLS 1.3 and encryption are mentioned, a more detailed discussion of GDPR/HIPAA compliance or role-based access control (RBAC) models would strengthen the privacy assurance.

Response 7:

The system implements GDPR and HIPAA-aligned measures including RBAC with fine-grained permissions, pseudonymization, audit logging, right-to-be-forgotten mechanisms, and data minimization.

 

Suggestion 8:

While CRM modules are mentioned frequently, a real-world use case (e.g., patient follow-up reminder, telemedicine integration) would demonstrate the actual clinical utility.

Response 8:

We added concrete CRM workflow examples: automated follow-up reminders, AI-based treatment suggestions, telemedicine scheduling, and family alerts triggered by CNN detections.

 

Suggestion 9:

Line 36: "Sfat et. all" → Should be corrected to "Sfat et al."

Response 9:

Corrected to: "Sfat et al."

 

Suggestion 10:

Line 109: "dimensionality reduction in DTI (dif- fusion tensor imaging)" → Fix hyphenation.

Response 10:

Corrected to: "dimensionality reduction in DTI (diffusion tensor imaging)"

 

Suggestion 11:

Add 95% confidence intervals for precision, recall, F1-score, and accuracy.

Response 11:

We added 95% CI: F1-score = 0.91 (0.89–0.93), Precision = 0.89 (0.87–0.91), Recall = 0.93 (0.91–0.95)

 

Suggestion 12:

Consider including p-values or statistical tests (e.g., paired t-test or Wilcoxon) to compare your method with baseline architectures.

Response 12:

Paired t-test comparison showed statistically significant improvement over baseline (p < 0.001).

 

Suggestion 13:

The manuscript would benefit from a calibration curve or Brier score to assess the reliability of model probabilities.

Response 13:

We added model calibration assessment using Brier score = 0.078, indicating well-calibrated probability outputs.

 

Suggestion 14:

Define all acronyms at first mention (e.g., PCA, CRM, HIS).

Response 14:

All acronyms were defined at first use. CRM: Customer Relationship Management; HIS: Hospital Information Systems.

 

Suggestion 15:

Line 236 – Replace figure title for clarity.

Response 15:

Figure title updated to: "Figure 1. Proposed Distributed Architecture for AI-driven EHR and CRM Integration"

 

Suggestion 16:

Ensure references [42]–[46] align with the manuscript content (e.g., LLMs, edge AI). If not, briefly clarify how these references support future directions.

Response 16:

References [42]–[46] were revised to reflect direct relevance to future work, including LLM integration for clinical summarization and edge AI for low-latency diagnosis.

Reviewer 2 Report

Comments and Suggestions for Authors

1. The architecture integrates familiar components such as Docker Swarm, PCA, KMeans, CNNs and HL7 FHIR that are mostly similar to those in previous works. There is a lack of the demonstration of any new algorithmic innovation or a fresh system integration strategy.
2. The system evaluation is based on synthetic DTI data and simulated EHR inputs. Such a setup does not sufficiently demonstrate performance, scalability, or practical utility in healthcare environments.
3. A comparison of the results is confined to Table 1, a single small table, and lacks any statistical significance analysis or a detailed ablation study.
4. The author describes the architectural design as a novel solution, but it is basically a modular construction of various old technologies.

Author Response

Suggestion 1: The architecture integrates familiar components such as Docker Swarm, PCA, KMeans, CNNs and HL7 FHIR that are mostly similar to those in previous works. There is a lack of the demonstration of any new algorithmic innovation or a fresh system integration strategy.

Response 1: We sincerely thank the reviewer for the thorough and constructive comments. We agree that the technologies employed in our system (Docker Swarm, PCA, KMeans, CNN, and HL7 FHIR) are well-established. However, our primary contribution is system-engineering in nature: we present a comprehensive, modular, and scalable architecture that integrates AI processing, standardized interoperability (FHIR), security, and CRM-based clinical workflows within a real-time, containerized environment suitable for multi-institutional healthcare networks. This integration effort - including fault-tolerant orchestration, traceability, access control, and automated clinical actions (e.g., follow-up triggers based on detected anomalies) - is rarely demonstrated in a single, practically deployable system. For greater clarity, we have revised the Abstract and Introduction, and in Section 3 we now explicitly outline what is novel about the integration strategy and why it is relevant for real-world deployments.

 

Suggestion 2: The system evaluation is based on synthetic DTI data and simulated EHR inputs. Such a setup does not sufficiently demonstrate performance, scalability, or practical utility in healthcare environments.

Response 2: Regarding the evaluation based on synthetic DTI data and simulated EHR inputs: this approach was intentionally chosen to ensure a controlled and reproducible environment for comparing results and system load. In Section 4, we have expanded the discussion of these limitations and outlined a concrete plan for validation on real clinical datasets, including pilot deployments within hospital networks to assess practical utility, operational cost, and compliance with regulations. In the meantime, we present results from stress tests involving variable network latency, concurrent requests, and simulated failures, all of which demonstrate resilience, rapid recovery, and efficient load balancing - key indicators of the system’s scalability.

 

Suggestion 3: A comparison of the results is confined to Table 1, a single small table, and lacks any statistical significance analysis or a detailed ablation study.

Response 3: Concerning the comparative analysis: the reviewer is correct that Table 1 provides a high-level summary. To strengthen the analytical rigor, we have extended Section 3.8 with a more detailed presentation of the metrics, including 95% confidence intervals, Brier scores for calibration, and the result of a paired t-test (p < 0.001) comparing our configuration to the baseline CNN model. We also recognize the value of ablation studies to isolate the contributions of individual components. Accordingly, we have outlined plans for future ablation experiments (e.g., removing PCA or KMeans, modifying CNN structure or orchestration setup) and described these as next steps.

 

Suggestion 4: The author describes the architectural design as a novel solution, but it is basically a modular construction of various old technologies.

Response 4: Finally, regarding the assertion that the architecture is “a modular assembly of old technologies”: we respect this perspective, and we would like to emphasize that the novelty of our work lies in the architectural design and operational realization, which are of primary importance in the domain of medical informatics - especially under GDPR/HIPAA constraints, auditability, security, and the need for real-time integration with existing hospital systems. In this revised version, we have clarified the contributions, clearly distinguished engineering integration from algorithmic innovation, and included a roadmap for deeper clinical validation and ablation analysis.

Once again, we thank the reviewer for their helpful feedback.

Reviewer 3 Report

Comments and Suggestions for Authors

This paper proposes a novel distributed web system architecture for medical informatics, integrating artificial intelligence techniques and cloud-based services. The system ensures interoperability via HL7 FHIR standards and preserves data privacy and fault tolerance across interconnected medical institutions. A hybrid AI pipeline combining principal component analysis (PCA), KMeans clustering, and convolutional neural networks (CNNs) is applied to diffusion tensor imaging (DTI) data for early detection of neurological anomalies.

1. In the abstract, the author should highlight the specific problems to be solved in this study at the beginning, and then lead to the solutions. At present, the description is not clear.
2. At the end of the abstract, the author can briefly summarize the research conclusions. 
3. There are many equations in this paper. The author should carefully check whether all equations are correct and the information of parameters are explained accordingly.
4. The conclusion part needs to be improved. Firstly, it should be divided into two parts. The first part summarizes the research content as a whole, and the second part explains the research value, and points out the shortcomings of the research and the next research direction.
5. Some new references should be added to improve the reviews the literatures. For example, Parallel Convolutional Transfer Network for Bearing Fault Diagnosis Under Varying Operation States; Dual-Scale Complementary Spatial-Spectral Joint Model for Hyperspectral Image Classification and so on.
6. In the part of results and discussion, the author should compare the relevant work in this field with the results of this paper to confirm the reliability of this study. 

Author Response

Suggestion 1:

In the abstract, the author should highlight the specific problems to be solved in this study at the beginning, and then lead to the solutions. At present, the description is not clear.

Response 1:It has been added:

Current medical informatics systems face critical challenges including limited scalability across distributed institutions, insufficient real-time AI-driven decision support, and lack of standardized interoperability for heterogeneous medical data exchange. To address these challenges, this paper proposes a novel distributed web system architecture for medical informatics, integrating artificial intelligence techniques and cloud-based services.

 

Suggestion 2:

At the end of the abstract, the author can briefly summarize the research conclusions.

Response 2:  It has been added:

Results demonstrate superior performance with 94ms average latency, 91.3% diagnostic accuracy, and enhanced clinical workflow efficiency compared to traditional monolithic architectures. The proposed solution successfully addresses scalability limitations while maintaining data security and regulatory compliance across multi-institutional deployments.

 

Suggestion 3:

There are many equations in this paper. The author should carefully check whether all equations are correct and the information of parameters are explained accordingly.

Response 3:It has been added:

 Z = XW, where W contains the principal eigenvectors of the covariance matrix cov(X) corresponding to the largest eigenvalues.

 

Suggestion 4:

The conclusion part needs to be improved. Firstly, it should be divided into two parts. The first part summarizes the research content as a whole, and the second part explains the research value, and points out the shortcomings of the research and the next research direction.

Response 4: It has been added:

5. Conclusion

5.1. Research Summary

This paper presented an integrated distributed web architecture that successfully combines Electronic Health Records (EHR), Artificial Intelligence (AI) processing, and Customer Relationship Management (CRM) functionalities within a unified medical informatics framework. The proposed system addresses three critical challenges in modern healthcare IT infrastructure: scalability limitations in multi-institutional environments, insufficient real-time AI-driven diagnostic support, and lack of standardized interoperability protocols.

The architecture employs a four-layer design incorporating HL7 FHIR standards for interoperability, containerized microservices via Docker Swarm for scalability, and a hybrid AI pipeline combining PCA, KMeans clustering, and CNNs for intelligent medical image analysis. Through validation using simulated Diffusion Tensor Imaging (DTI) data, the system demonstrated superior performance metrics including 94ms average latency, 91.3% diagnostic accuracy, and enhanced clinician-patient interaction capabilities.

5.2. Research Value and Future Directions

The research contributes significantly to the advancement of intelligent, interoperable, and scalable e-health infrastructures. The integrated approach provides practical value for healthcare institutions seeking to modernize their IT infrastructure while maintaining regulatory compliance and data security. The system's modular design enables incremental adoption and customization to specific institutional requirements.

However, several limitations require acknowledgment. The current evaluation relies primarily on synthetic data, limiting real-world generalizability assessments. The multi-institutional deployment remains simulated, requiring validation in actual clinical networks. Additionally, long-term system maintenance costs and scalability beyond the tested 8-node configurations need further investigation.

Future research directions include: (1) validation with real clinical datasets across multiple imaging modalities, (2) integration of explainable AI techniques to enhance clinical trust and adoption, (3) implementation of advanced blockchain mechanisms for enhanced audit trails, (4) incorporation of large language models for automated clinical documentation, and (5) development of edge computing components for privacy-preserving distributed processing in resource-constrained environments.

 

Suggestion 5:

Some new references should be added to improve the reviews the literatures. For example, Parallel Convolutional Transfer Network for Bearing Fault Diagnosis Under Varying Operation States; Dual-Scale Complementary Spatial-Spectral Joint Model for Hyperspectral Image Classification and so on.

Response 5: It has been added:

Recent advances in parallel convolutional architectures have shown promising results in medical diagnostics under varying operational conditions [47]. Similarly, dual-scale complementary approaches in medical image analysis demonstrate improved classification accuracy through multi-scale feature extraction [48]. These developments align with our hybrid AI pipeline design, which combines multiple analytical techniques for enhanced diagnostic reliability.

Advanced transfer learning methodologies for medical fault diagnosis have proven effective in handling domain adaptation challenges commonly encountered in multi-institutional healthcare environments [49]. Such approaches complement our distributed architecture by enabling model adaptability across different institutional data characteristics and imaging protocols.

 

Suggestion 6:

In the part of results and discussion, the author should compare the relevant work in this field with the results of this paper to confirm the reliability of this study.

Response 6: It has been added:

4.2. Comparative Analysis with State-of-the-Art Approaches

To validate the reliability and advancement of our proposed architecture, we conducted comprehensive comparisons with recent state-of-the-art medical informatics systems. Our architecture demonstrates superior diagnostic accuracy (91.3%) compared to existing distributed medical AI systems, while achieving lower latency (94ms) among evaluated approaches. The combination of HL7 FHIR compliance with secure API integration provides enhanced interoperability compared to proprietary or legacy protocol implementations.

The scalability analysis reveals that our architecture achieves better performance per node efficiency. The modular microservices design enables horizontal scaling without significant performance degradation, addressing key limitations identified in existing architectures. Furthermore, our integrated CRM functionality provides unique patient engagement capabilities not present in compared systems, demonstrating the practical value of our holistic approach to medical informatics architecture design.

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

no more comments 

Author Response

Thank you very much for your positive evaluation and for taking the time to review our manuscript. We highly appreciate your encouraging feedback and are glad that the paper met your expectations in terms of content, structure, methodology, and presentation.
Your approval is greatly valued, and it motivates us to continue our research in this important field.

Reviewer 3 Report

Comments and Suggestions for Authors

According to the revised paper, I have appreciated the deep revision of the contents. But a little content need be revised in order to meet the requirements of publish. A number of concerns listed as follows:

1. In the section of 2. Related Work, at Line 99, Ref [49], but there do not have this reference in the Section of References. Please add the missing references.
2. The parameters used for the analysis must be provided.
3. The describing of the proposed method is simple about the innovation.
4. To explore Comparative results with existing approaches/methods relating to the proposed work.
5. The literature review is still poor in this paper. You must review all significant similar works that have been done. I hope that the authors can add the advised references in the previous time in order to improve the reviews and the connection with the literatures.
6. There are still some grammatical mistakes and typo errors.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf