Secure Federated Learning for Privacy‑Preserving Medical Record

Sharing across Hospitals

Sivakumar Ponnusamy

, R. Mohemmed Yousuf

, G. Visalaxi

, S. Sumithra

B. Sushma

and Shree Yoghitha S.

Department of Computer Science and Engineering, K.S.R. College of Engineering, Tiruchengode, Namakkal, Tamil Nadu,

India

Department of Information Technology, Bannari Amman Institute of Technology, Erode, Tamil Nadu, India

Department of CSE, S.A. Engineering College, Chennai, Tamil Nadu, India

Department of Electronics and Communication Engineering, J.J. College of Engineering and Technology, Tiruchirappalli,

Tamil Nadu, India

Department of Information Technology, MLR Institute of Technology, Hyderabad, Telangana, India

Department of MCA, New Prince Shri Bhavani College of Engineering and Technology, Chennai, Tamil Nadu, India

Keywords: Federated Learning, Privacy‑Preserving, Medical Records, Secure Data Sharing, Healthcare Institutions.

Abstract: In the rapidly-changing era of eHealth, secure and effective inter-institutional data sharing is still a major

issue. In this work, we introduce a federated learning method for privacy-preserving medical record sharing

between a collection of hospitals. With the ability to train models in a decentralized manner, all without

moving aggregated sensitive patient data, the proposed method respects data sovereignty, complies with

regulation and promotes collective intelligence. The design includes efficient encryption and lightweight

communication algorithms, so in secure way computational overhead is optimally minimized. The utility is

evaluated with a variety of hospital databases and shows excellent performance in terms of data privacy, the

diagnostic accuracy, and the interoperability. This work not only addresses the restrictions of centralized data

systems, but creates a foundation for scalable, secure, and responsible AI deployment in health systems.

1 INTRODUCTION

Breakneck digital transformation in healthcare has

resulted in the creation of vast amounts of highly-

sensitive patient data across organizations. Now that

the hospital systems, healthcare groups and clinical

personnel have adopted intelligent systems for

diagnosis, treatment recommendation and patient

surveillance the necessity for cooperative data

integration and use has ever been more important.

Yet, the proposed centralized schemes for

aggregating data is not without significant drawbacks

that include patient privacy, security, privacy

regulations like HIPAA and GDPR. Many of the

traditional machine learning algorithms depend on

data being centralized collected in a single location,

opening it up to a central point of failure, which is

the data centre itself.

Federated learning (FL) has been recognized as a

promising solution to overcome the above challenges

through decentralized model training without sharing

raw data. This strategy prescribes that models are

trained on the local datasets of institutions and that

only the model updates are confirmed in the seeding

of a percent of the global model. This not only

ensures data privacy but also supports collective

wisdom among geographically remote healthcare

systems. It’s one big challenge is to overcome

communication overhead, model convergence, and

data heterogeneity in existing FL frameworks,

especially in a real-time clinical environment.

This research presents a novel FL framework for

sharing medical records securely and privately. The

system incorporates strong cryptography schemes,

effective communication strategies, and progressive

learning algorithms to improve the security and

performance issues. In contrast to prior work, the

proposed system is evaluated over various health care

datasets, tested in terms of usability, scalability, and

real-time performance. By addressing the privacy-

194

Ponnusamy, S., Yousuf, R. M., Visalaxi, G., Sumithra, S., Sushma, B. and S., S. Y.

Secure Federated Learning for Privacy-Preserving Medical Record Sharing across Hospitals.

DOI: 10.5220/0013860100004919

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 1st International Conference on Research and Development in Information, Communication, and Computing Technologies (ICRDICCT‘25 2025) - Volume 1, pages

194-201

ISBN: 978-989-758-777-1

preserving guarantees of collaborative data analysis,

it bridges the gap between health data privacy

protection and collaborative analytical research for

practical and scalable promotion of secure and ethical

AI-driven healthcare.

2 PROBLEM STATEMENT

The healthcare sector is tackling the tough task of

ensuring personal patient medical records can be

securely shared between institutions transparently

while keeping all personal data private, compliant

with regulations, and interoperable. Conventional

centralized ML methodologies depend on data

syndication leading to risk of breach of data and

privacy law infringements. However, while federated

learning is appealing due to its decentralized solution,

existing federated optimization frameworks often

lack realistic adaption to real clinical settings, in

terms of communication efficiency, accuracy on non-

IID data of the learned model, and seamless

integration into heterogeneous healthcare systems. A

federated learning framework is urgently required

that is designed for secure, privacy-preserving, and

efficient data sharing on medical data contribution

from different healthcare institutions without data

ownership and diagnostic accuracy being affected.

3 LITERATURE SURVEY

Driven by the increasing need of secure and

collaborative healthcare analytics, the study of

federated learning (FL) as an alternative to central

data processing has gained momentum. Li et al.

(2021) also stressed the significance of privacy-first

research that utilizes FL for medical applications,

where holding the ability to train models without

sharing any sensitive patient data stands out.

Froelicher et al. (2021) applied multiparty

homomorphic encryption in FL for better security and

privacy in precision medicine. Similarly, Che et al.

(2021) addressed federated multi-view learning in the

context of private medical data integration, and

developed new models that uphold data locality and

enable collaborative intelligence.

Recent studies, including those by Nguyen et al.

(2021) and Ali et al. (2024), has done thorough survey

of FL in smart healthcare including definition, work

concept, typical application scenarios, and so on

along with security threats faced by FL-based smart

healthcare applications, difficulties for security, and

development direction. In contrast, these reviews also

raise concerns about the practical use of FL based on

issues of data heterogeneity, communication cost, and

latency. Muthalakshmi et al. (2024) resolved these

concerns to some extent by proposing a decentralized

FL model for medical image analysis, and Xiang et

al. (2025), designed for privacy-preserving

collaborative analysis on EHR with federated

approach.

The direct secureization and lightness

improvements on FL are also remarkable. Kumar et

al. (2019); however, the complexity added the system

overhead when blockchain technology was combined

with FL to ensure medical image sharing (Parizi et

al., 2021). To mitigate these limitations, Zhou et al.

(2025) introduced a performance-optimized FL

frame for EHR analysis less computationally

confounded. Sav et al. (2021) presented POSEIDON,

a privacy-preserving neural network learning

framework, demonstrating secure training with little

data exposure.

The question of interoperability and integration of

the data across institutions persist. Scheibner et al.

(2021) called for stronger privacy-enhancing

technologies that would allow the ethical and legal

sharing of medical data. Similarly, Zhang et al. (2023)

created FL-enabled analytics applications, which are

targeted to support private yet large-scale eHealth

studies. Rodrguez-Barroso et al. (2020) considered

privacy-respecting software frameworks like Sherpa.

ai, in the context of their application in medicine.

A wider industry perspective is also reflected in

the sector. Commercial efforts such as those

described in (Owkin et al.2023), (Vinluan 2022), and

(Wiggers 2020) demonstrate the real-world appetite

and spending for federated models across both drug

discovery and diagnostic intelligence. On the other

hand, existing work such as Yang et al. (2019) and

Shokri et al. (2011) that give a theoretical foundation

for federated learning and privacy quantification,

respectively, on which all modern FL libraries build.

Although previous studies made legitimate

ground work, for the development of adaptable,

scalable, and performing federative systems, for a

large and widely varying medical infrastructure,

challenges still exist. In this work, we attempt to

move the needle on such disjunction by proposing a

federated learning technical solution aiming for

privacy of the patients, low-latency communication,

secure sharing of medical records across institutions

by featuring a federated setup that we shall refer to as

Federated Medical Learning Framework (FerMLF).

Secure Federated Learning for Privacy-Preserving Medical Record Sharing across Hospitals

195

4 METHODOLOGY

The federated learning-based secure and privacy-

preserving medical records sharing scheme among

healthcare institutions is designed to achieve the

sharing and utilization of multi-center medical data

while ensuring both privacy and model performance.

This approach is intended to address the fundamental

flaws of centralized solutions - the shortcomings of

managing data, regulatory issues and poor

communication between systems. The combined

federated learning with encryption, Communication-

Efficient Protocols and adaptive learning enables the

data-privacy reservation, latency reduction and

accurate collaborative learning.

Figure 1: Federated Learning Workflow for Secure Medical

Record Sharing.

The first stage of the proposed method is to

collect data from multiple hospitals. Each

participating institution is in control of all its

downloaded data, such as EMRs, imaging, and

sensor data. This data is never shared among

institutions, but is used for training local models

directly at the point of care without violating patient-

level privacy. Each hospital has access to its own

medical records, and utilizes these data to train local

machine learning models on their specific patient

population. Such a set up keeps patient data in the

school’s sphere at all times and conforms to the

relevant privacy standards, such as HIPAA and

GDPR. Figure 1 shows the Federated Learning

Workflow for Secure Medical Record Sharing.

The proposed methodology has the first stage of

data gathering from more than one hospital. Each

participating organization is responsible for their

local data, which including electronic medical records

(EMRs), diagnostic images, and sensor data. This

data is never exchanged among institutions, but is

exploited for training local models at the point of care

without compromising individual patient privacy.

Each hospital has access only to its own medical

records and employs them to train local machine

learning models that are specific to the patient

population of that hospital. This arrangement means

that patient information is always within the school’s

domain and adheres to appropriate privacy standards,

like HIPAA and GDPR.

When the local models are trained, they are

aggregated. While in federated learning, model

aggregation sends only the model parameters or

updates instead of raw data to a central server for

global model training. With this method, the

aggregation authority never sees patients’ actual

detailed data and patient privacy is preserved. The

updates at each participating institution are sent to a

central server, which combines updates across the

institutions to produce a global model enhanced with

the experience of the entire network without ever

seeing or learning individual patient data. Model

aggregation is performed by methods like Federated

Averaging, which averages the model updates from

each contributing organ weighted by the numbers of

subjects from that contributing organ.

The encryption methods are important for the

privacy of the model update in the aggregation

process. Homomorphic encryption, which is capable

of performing computation on encrypted data

without revealing the data, is used to add more

privacy for aggregated data. This encryption

guarantees that the central server never has access to

the raw model parameters, and only ever receives

encrypted updates, to avoid any possible data

leakage. Moreover, low computational proficient

algorithms are adopted to optimize a homomorphic

encryption and decryption, making it practical for

real-time medical systems.

Communication overhead is another challenge,

i.e., the bandwidth overhead of transmitting model

updates between institutions and central server as the

model communicates with a central server in

federated learning. To solve this, the framework

includes optimized communication protocols which

minimize data exchange size and frequency. For

instance, by transmitting only a large value, the

amount of unnecessary transmission is reduced.

Furthermore, we deploy adaptive strategies which can

vary the updating frequency according to the

convergence speed of the model. ” These

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

196

approaches allow institutions to train models without

relying too much on network infrastructure such as in

environments with low bandwidth and/or real-time

demands.

Another main aspect of the methodology is how

non-IID data among institutions is managed. In

health care domain, medical data differs naturally

between different institutions because of different

patient population, medical procedures and testing

methods. To solve this problem, the framework

integrates methods to enhance the model

convergence in non-IID data. One such approach is

Personalized Federated Learning, which consists in

making the model of each institution encode the local

data of the institution while still contributing to the

global model. This personalisation guarantees that the

global model maintains good performance on a wide

range of datasets and medical situations.

It also covers model testing and feedback loops in

a live environment. Training and aggregation

continue until global model is distributed itself back

to the institutions to make real-time predictions and

decisions. All its readings are being recorded on the

fly and sent back to the central server that can

compensate for local data distributions or local

accuracy performance. The model is refined by using

the model performance in clinical practice as

feedback for improvement. This cyclic process

enables the model to reflect changing healthcare

environment and evolving patient demographics over

time.

So that ethical and privacy aspects are considered

and made transparent, the architecture provides an

audit mechanism that tracks the flow of data and the

respective model updating. The audit trail records

every mail exchange between the institutions and the

central server as well as when and what model was

updated and what the parameters for the model was

changed to. This auditing process makes the usage of

federated learning for sharing medical data

transparent and accountable, which is crucial for user

trust in the system.

Finally, we test the proposed LLDQ on real

healthcare data to show the scalability, the security

and the efficiency of the method. These assessments

are performed by testing the system in a certain

number of centres that include their training medical

records to train a global model. The framework is

evaluated with respect to model accuracy,

communication cost, training time, and data privacy

in varying scenarios. The results of this series of

analyses is compared with the combination of

federated learning generations so to highlight the

progress of data privacy, model efficacy, and system

efficiency.

In summary, this paper mainly focuses on secure

and privacy-preserving medical-record sharing, and

the challenge of high performance and scalability.

With the aid of decentralized training, homomorphic

encryption based secure data sharing,

communication-efficient optimization algorithms,

and adaptive learning, the proposed system is a

practical solution for collaborative healthcare

analytics across organizations. This not only

addresses the issue of data privacy and security, but

also could potentially make machine learning more

powerful, more effective, when it is applied in

medicine, providing ways for secure, scalable AI-

driven healthcare.

5 RESULTS AND DISCUSSION

Empirical evaluation has been achieved to indicate

that the proposed FL framework for secure and

privacy-preserving medical records sharing among

health-care institutions is able to provide effective

data privacy, model performance and system

scalability through experimental results. Several

hospitals were included in the experiment and had

information added to their dataset to help them train

a decentralized model. Experimental results

demonstrate that our framework effectively handles

the issue of raw input privacy, communication

overhead and model convergence when applied to the

non-IID nature of medical data while preserving high

accuracy under the common learning environment

among heterogeneous hospitals.

Figure 2: Model Accuracy Improvement With Increasing

Institutions.

Secure Federated Learning for Privacy-Preserving Medical Record Sharing across Hospitals

197

Table 1: Model Performance Metrics.

Metric

Description

Federated Model

Baseline Model

Accuracy

The percentage of correct predictions

92%

85%

Precision

The ability to avoid false positives

90%

83%

Recall

The ability to identify all positive cases

88%

80%

Communication

Overhead

Bandwidth used during model updates

50 KB per update

200 KB per update

Privacy Preservation

Compliance with data privacy

regulations

HIPAA, GDPR

N/A

Privacy on enter data while on training was one of

the main concerns of the framework. The federated

learning model is superior to the centralized model, in

which the data are sent to or stored in a centralized

place, which is beneficial for the privacy

preservation because only the data is saved in

individual institutions. This was made possible by

incorporating encrypted protocols (e.g.,

homomorphic encryption) to ensure that the update

during aggregation was secure. The test also

demonstrated that the raw patients’ data was not

transmitted to the central server, but only the

encrypted updates of the model were shared. This

methodology ensured that the patient data remained

secure, and privacy laws such as the HIPAA and the

GDP friendly, as no sensitive data was ever passed to

the cloud vendors. Furthermore, the total audit trail

managed by the framework offered proof of model

updates, making it a transparent and traceable means

for the data to be shared. Figure 2 shows the Model

Accuracy Improvement with Increasing Institutions. Table

1 shows the Model Performance Metrics.

Table 2: Real-Time Model Evaluation Results.

Institution

Model Accuracy

Precision

Recall

Feedback Rating

Institution A

94%

92%

89%

Positive

Institution B

90%

88%

85%

Neutral

Institution C

93%

91%

90%

Positive

Institution D

91%

89%

87%

Positive

From a modelling perspective, the federated

learning scheme showed a great improvement on the

diagnostic accuracy in various medical fields. Two

medical datasets such as healthcare datasets, both

clinical data and medical images, and sensor data,

were evaluated showing the superiority of the global

model over baseline models trained by centralised

data. The federated model accuracy was stable across

institutions, although their locally data distributions

were different. This abundance of personalized

federated learning may have properly adjusted local

model to institution-specific data without sacrificing

contribution to the global model. The generalization

over various datasets was indeed critical to afford

that the global model was still robust and accurate

even after training on non-IID data. Table 2 shows the

Real-Time Model Evaluation Results.

Table 3: Federated Learning Performance With Varying Number of Institutions.

Number of

Institutions

Model Accuracy

Communication Overhead

Latency

Training Time

(hrs)

90%

100 KB per update

2 secs

92%

150 KB per update

3 secs

94%

250 KB per update

5 secs

95%

400 KB per update

7 secs

95.5%

500 KB per update

10 secs

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

198

Another more attractive component of the

evaluation was the decrease in communication

overhead. Communication in federated learning tends

to be expensive. The latter two would dominate the

bandwidth usage but by the use of optimized

communication protocols, data transmission was

minimized greatly by the developed approach.

Updates in the model were sent back to the central

server and excess communication was reduced.

Furthermore, an adaptive learning schedule was

employed to adapt the update frequency according to

the model convergence. With this method, bandwidth

and latency were minimized, while the system could

remain efficient and responsive for application in

real-time medical operations. These optimizations

were especially critical in resource-constrained

environments. Table 3 shows the Federated Learning

Performance with Varying Number of Institutions.

Figure 3: Communication Overhead With Increasing

Institutions.

Another aspect that we wanted to evaluate is the

capability of the framework to handle non-IID data.

Healthcare for medical data is very heterogeneous,

too, from one hospital to another hospital because of

different demographic, same of medical procedures,

and diagnostic technology. The diversity of data has

posed a challenge for the federated learning

paradigm, and the convergence for non-IID data is

hard to be obtained based on conventional algorithms.

The experiments demonstrated the effectiveness of

processing this diversity of data using personalized

learning models in the federated learning setting. The

performance of the system was higher than that of

conventional FL models in which data is considered

IID and all of the data is shared, thereby allowing

institutions to adapt their local models to better fit

their data. This adaptation allowed both models to

converge and the general model to be applied to the

specifics of the single institutions. Figure 3 shows

the Communication Overhead with Increasing

Institutions.

In addition, the end-to-end platform was

evaluated in a multi-hospital environment. It was

shown that the system can be easily expanded to

include new institutes without a significant drop in

model performance and communication efficiency.

Even when more scanning institutes are added, the

global functional model worked well and the

communication cost was feasible. This scalability is a

key consideration for the future introduction of the

solution on a large scale, thus the solution can be

scaled up for a larger number of providers and yet

performance and security are maintained.

Figure 4: Latency Impact With Increasing Institutions.

However, there were some negative sides noticed

during the testing – yet very few. The expensive

encryption of cryptographic processor, especially in

real time mode, was viewed as one of the main

drawbacks. While the use of homomorphic

encryption enabled privacy, it resulted in model

aggregation timeouts, particularly on large datasets.

To address these, a more efficient encryption scheme

can be employed or any resultant homomorphic

encryption can be tuned to reduce computational

overhead. In addition, while this provides a step

forward, a personalized federated learning model

still performed poorly in the presence of imbalance

data distribution, where the model converged at a

slower rate. New more efficient aggregation method

can be explored in the future to alleviate these

overheads. Figure 4 shows the Latency Impact with

Increasing Institutions.

Figure 5: Privacy Compliance in Federated Learning.

Secure Federated Learning for Privacy-Preserving Medical Record Sharing across Hospitals

199

In summary, our experiments show that the

proposed federated learning framework provides a

secure, privacy-preserving sharing of medical records

between healthcare institutions. The system

effectively solved the main problems, including data

privacy, communication overhead and model

performance, with accurate and scalable predictions.

There are still some limitations that need to be

addressed to as above-mentioned, such as optimizing

encryption algorithms, handling highly skewed

distributions of data; however, it is high time for

realizing safe, privacy-preserving, robust and

intelligent healthcare analytics for the future. This

work adds to the emerging literature on federated

learning for healthcare and provides a basis for

further development of secure and collaborative

sharing of medical data. Figure 5 shows the Privacy

Compliance in Federated Learning.

6 CONCLUSIONS

The rising demand of secure, privacy preserving

EMR sharing among multiple healthcare providers is

a critical issue traditional, centralized system is

difficult to meet. The federated learning paradigm

introduced in new study gives a decentralized way to

enable cooperation in model learning and preserve

privacy and security for patient sensitive data. By

leveraging strong encryption methods, efficient

communication protocols and personalized learning

techniques, the framework effectively mitigates the

privacy concerns related to sharing the data, while

each participating institution retains full control of its

local data.

Federated learning provided a practically feasible

solution to fit for data privacy concerns,

communication overhead and model performance, as

evaluated in the framework. Because the framework

keeps the sensitive personal medical data locally in

institutions and only sends the model updates, the

privacy of patients is protected and institutions can

cooperate on improving diagnostic accuracy. The

application of personalized federated learning

models further enhanced the capability for the system

to cope with the heterogeneity from medical data

across institutes and to keep the global model robust

and accurate with data distribution variations.

However, there are some remaining challenges, as

the optimisation of encryption algorithms on real-

time signals and the data unbalance and model

convergence for highly imbalanced data. More

efficient encryption and better aggregation

techniques could be studied to address these

limitations in the future. Upon resolving these

challenges, the proposed framework offers a scalable,

secure, and practical federated learning solution for

healthcare, and serves as an enabler for AI-driven

healthcare analytics across institutions.

This study provides an important reference for

the direction of federated learning in the field of

healthcare, as it is shown that the privacy and security

of data can be well protected while realizing

collaborative intelligence. Through improved

understanding and adoption of federated learning this

work contributes to the realization of more secure and

scalable healthcare systems, that ultimately can result

in better patient outcome and cost -effective medical

practices.

In conclusion, the results demonstrate that the

proposed federated learning framework offers a

robust solution for secure, privacy-preserving sharing

of medical records across healthcare institutions. The

framework successfully addressed the primary

challenges of data privacy, communication overhead,

and model performance, while maintaining high

accuracy and scalability. Although there are still areas

for improvement, particularly in optimizing

encryption techniques and addressing highly skewed

data distributions, the framework presents a

significant step forward in enabling secure,

decentralized AI-driven healthcare analytics. The

findings from this study contribute to the growing

body of research on federated learning in healthcare

and pave the way for future advancements in secure

and collaborative medical data sharing.

REFERENCES

Abu El Houda, M., et al. (2022). Decentralized federated

learning model for robust and privacy-preserving

gradient aggregation. Knowledge-Based Systems, 250,

108927. PMC

Ali, M. S., Ahsan, M. M., Tasnim, L., & Ahmed, M. M.

(2024). Federated learning in healthcare: Model

misconducts, security, challenges, applications, and

future research directions—A systematic review. arXiv

preprint arXiv:2405.13832. arXiv

Che, S., Peng, H., Sun, L., Chen, Y., & He, L. (2021).

Federated multi-view learning for private medical data

integration and analysis. arXiv preprint

arXiv:2105.01603. arXiv

Froelicher, D., Troncoso-Pastoriza, J. R., Raisaro, J. L.,

Cuendet, M. A., & Sa Sousa, J. (2021). Truly privacy-

preserving federated analytics for precision medicine

with multiparty homomorphic encryption. Nature

Communications, 12(1), 1-10. Wikipedia

Hubaux, J. P., et al. (2021). Secure sharing of health data.

IEEE Euro S&P 2020 Keynote Speech.Wikipedia

ICRDICCT‘25 2025 - INTERNATIONAL CONFERENCE ON RESEARCH AND DEVELOPMENT IN INFORMATION,

COMMUNICATION, AND COMPUTING TECHNOLOGIES

200

Kumar, R., Wang, W., Yuan, C., Kumar, J., Zakria, H. Q.,

Yang, T., & Khan, A. A. (2021). Blockchain-based

privacy-preserved federated learning for medical

images: A case study of COVID-19 CT scans. arXiv

preprint arXiv:2104.10903. arXiv

Li, W., et al. (2021). Privacy-first health research with

federated learning. npj Digital Medicine, 4(1), 1-5.

Nature+1Wikipedia+1

Muthalakshmi, M., Jeyapal, K., Vinoth, M., & Dinesh, P.

S. (2024). Federated learning for secure and privacy-

preserving medical image analysis in decentralized

healthcare systems. In 2024 5th International

Conference on Electronics and Sustainable

Communication Systems (ICESC) (pp. 1-6). IEEE.

ResearchGate

Nguyen, D. C., Pham, Q. V., Pathirana, P. N., Ding, M.,

Seneviratne, A., Lin, Z., ... & Hwang, W. J. (2021).

Federated learning for smart healthcare: A survey.

arXiv preprint arXiv:2111.08834.

Owkin. (2023). Federated learning in healthcare:

Preserving privacy, unleashing potential.

ResearchGate.

Rani, S., Kataria, A., Kumar, S., & Tiwari, P. (2023).

Federated learning for secure IoMT-applications in

smart healthcare systems: A comprehensive review.

Knowledge-Based Systems, 250, 108927.

Rodríguez-Barroso, N., Stipcich, G., Jiménez-López, D.,

Ruiz-Millán, J. A., & Martínez-Cámara, E. (2020).

Federated learning and differential privacy: Software

tools analysis, the Sherpa.ai FL framework and

methodological guidelines for preserving data privacy.

Information Fusion, 64, 297-306.

Sav, S., Pyrgelis, A., Troncoso-Pastoriza, J. R., Froelicher,

D., & Bossuat, J. P. (2021). POSEIDON: Privacy-

preserving federated neural network learning. In

Proceedings of the 2021 Network and Distributed

System Security Symposium.

Scheibner, J., Raisaro, J. L., Troncoso-Pastoriza, J. R.,

Ienca, M., & Fellay, J. (2021). Revolutionizing medical

data sharing using advanced privacy-enhancing

technologies: Technical, legal, and ethical synthesis.

Journal of Medical Internet Research, 23(2), e25120.

Shokri, R., Theodorakopoulos, G., Le Boudec, J. Y., &

Hubaux, J. P. (2011). Quantifying location privacy. In

2011 IEEE Symposium on Security and Privacy (pp.

247-262). IEEE.

Trinanda Putra, K., Chen, H. C., Shyu, C. R., & Tsai, Y. Y.

(2021). A systematic review of federated learning in the

healthcare area: From the perspective of data properties

and applications. Applied Sciences, 11(11), 5173.

Vinluan, F. (2022). Sanofi exec jumps to Owkin to ramp up

the AI biotech's pharma partnership plans. BioPharma

Dive. Wikipedia

Wiggers, K. (2020). Major pharma companies, including

Novartis and Merck, build federated learning platform

for drug discovery. VentureBeat. Wikipedia

Xiang, Y., et al. (2025). Privacy-preserving federated

learning for collaborative medical data analysis.

Scientific Reports, 15(1), 1-12.PMC+1Nature+1

Yang, Q., Liu, Y., Chen, T., & Tong, Y. (2019). Federated

machine learning: Concept and applications. ACM

Transactions on Intelligent Systems and Technology

(TIST), 10(2), 1-19.

Zhang, Y., et al. (2023). Advancing privacy-preserving

health care analytics and research with federated

learning. JMIR AI, 2(1), e60847. JAI - JMIR AI

Zhou, Y., et al. (2025). Leveraging federated learning for

privacy-preserving analysis of electronic health

records. Artificial Intelligence in Medicine, 135, 102

Secure Federated Learning for Privacy-Preserving Medical Record Sharing across Hospitals

201