Enhancing Healthcare Data Privacy and Security Using Federated Learning, Encryption and Privacy-Preserving Record Linkage (PPRL)
Open Access DepositedAs the healthcare data-sharing landscape evolves, privacy-preserving collaborative use of sensitive medical data presents critical issues. In this praxis, we introduce a hybrid framework for Federated Learning (FL), which combines Horizontal Federated Learning (HFL) and Vertical Federated Learning (VFL) techniques along with the use of Federated Averaging (FedAvg), AES-256 encryption, and Privacy-Preserving Record Linkage (PPRL) to address concerns for data security and privacy. The author utilizes the publicly available pediatric chest X-ray dataset from Kaggle for this work. Thus, this study emulates a realistic multi-institutional healthcare setting where image data and associated metadata are distributed across multiple institutions/sites. AES-256 encryption provides security for the parameters of the model and communication between clients, whereas PPRL, based on a Bloom filter, allows patients’ records to be linked in a secure way (Vatsalan et al., 2020).The hybrid framework is capable of training convolutional neural networks (CNNs) and multilayer perceptrons (MLPs) at partitioned nodes while keeping raw patient data hidden. To enhance generalizability, separate HFL and VFL pipelines were independently trained and later merged using FedAvg. Our experiment achieved high F1 scores with both the HFL model and the VFL model. The final model was able to generalize well on unseen data from reference clients. Still, because data distributions are heterogeneous in healthcare systems, the aggregated model experienced a slight drop in accuracy, which is commonly observed in privacy-preserving collaborative models. To sum up, the results demonstrate that the proposed hybrid FL architecture can be a useful and scalable means of deploying artificial intelligence in healthcare. This praxis employs AES encryption for secure communication, Bloom filter-based linkage for anonymized identity resolution, and Federated Learning for decentralized training. Collectively, it offers a reproducible benchmark for future health analytics. It sets the stage for applying FL frameworks in situations found in actual clinical settings where privacy, interoperability, and model robustness are all critical.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.
| Thumbnail | Title | Date Uploaded | Visibility | Actions |
|---|---|---|---|---|
|
|
Jatavallabhula_gwu_0075A_17521.pdf | 2025-12-11 | Open Access |
|