EUVIMEDEUVIMED entdecken
Crossref

Privacy‐Preserving Federated Multimodal Deep Learning for Sepsis Prediction Using Vision Transformers and Secure Aggregation

Bhanu Prakash Reddy Rella, Arvind Kumar Chaudhary, Ronish B. Patel, Bosko Nikolic, Milos Janjic, Nebojsa Bacanin

Internet Technology Letters · 2026-07 · 9 (4)

DOI: 10.1002/itl2.70324 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Internet Technology Letters, online erschienen am 08.06.2026. Verfasst von Bhanu Prakash Reddy Rella und 5 weiteren Autor:innen. Die Arbeit nennt 29 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Multimodal clinical data such as medical imaging, electronic health records (EHRs) and genomic information have become increasingly important for intelligent healthcare analytics and early disease prediction. However, centralized AI training approaches introduce major concerns related to patient confidentiality, institutional data governance, and secure interhospital collaboration. To address these limitations, this work presents a privacy‐aware federated multimodal learning framework for sepsis prediction in distributed healthcare environments. The proposed architecture combines a Vision Transformer (ViT) for chest X‐ray feature extraction with a Deep Neural Network (DNN) for structured EHR and genomic data processing, enabling efficient multimodal feature fusion without centralized data sharing. Differential Privacy is incorporated to protect local model updates, while Homomorphic Encryption enables secure aggregation during federated communication. The framework was evaluated using multimodal clinical datasets containing chest radiographs, patient records and genomic indicators collected across simulated healthcare institutions. Experimental findings demonstrate that the proposed framework achieves an AUC of 0.945 and an F1‐score of 0.887 while maintaining strong privacy guarantees and robustness against inference attacks. Comparative analysis further shows that the proposed method achieves performance close to centralized learning while significantly improving data confidentiality and collaborative security. The framework provides a scalable and practical solution for privacy‐preserving healthcare AI deployment across distributed medical systems.

Zitieren

APA

Rella, B. P. R., Chaudhary, A. K., Patel, R. B., Nikolic, B., Janjic, M., & Bacanin, N. (2026). Privacy‐Preserving Federated Multimodal Deep Learning for Sepsis Prediction Using Vision Transformers and Secure Aggregation. Internet Technology Letters, 9(4). https://doi.org/10.1002/itl2.70324

Vancouver

Rella BPR, Chaudhary AK, Patel RB, Nikolic B, Janjic M, Bacanin N. Privacy‐Preserving Federated Multimodal Deep Learning for Sepsis Prediction Using Vision Transformers and Secure Aggregation. Internet Technology Letters. 2026;9(4). doi:10.1002/itl2.70324

BibTeX
@article{Rella2026,
  author = {Rella, Bhanu Prakash Reddy and Chaudhary, Arvind Kumar and Patel, Ronish B. and Nikolic, Bosko and Janjic, Milos and Bacanin, Nebojsa},
  title = {Privacy‐Preserving Federated Multimodal Deep Learning for Sepsis Prediction Using Vision Transformers and Secure Aggregation},
  journal = {Internet Technology Letters},
  year = {2026},
  volume = {9},
  number = {4},
  doi = {10.1002/itl2.70324}
}
Kostenlos starten →