Fairfedtransnet: Fairness-Aware Transformer-Based Federated Learning for Multimodal Rare Disease Diagnosis

View at publisher Summarise with VIRAT
Authors: Bhavana Jamalpur, KENNEDY JEEVARATHINAM, K. Vetrivel, S.Sesha Vidhya, M. Ashok, B. Jegajothi
Cited by: 0
References: 16

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Diagnosis of rare diseases is a pressing issue because of the small prevalence and the multimodal nature of the clinical information and the rigid privacy demands across institutions. Federated learning (FL) allows collaborative learning without centralizing sensitive data, yet currently, the strategies used do not take into account fairness among clients, multimodality, and adaptability to the real-world deployment. We introduce FairFedTransNet a fairness-conscious transformer-based FL model with Multi-Objective Evolutionary Algorithms (MOEA). The framework includes: (i) a dual-stream transformer encoder of structured and temporal EHR data, (ii) cross-modal fusion of EHR, imaging, and genomic signals through attention-guided alignment, (iii) fairness-conscious adaptive aggregation to reduce client disparities, and (iv) MOEA-based optimization to optimize the combination of accuracy and fairness with the control of hyperparameters and aggregation weights. An adaptive differential privacy mechanism further safeguards sensitive gradients with minimal performance trade-off. Thorough experiments using the multi-institutional eICU dataset indicate that FairFedTransNet can obtain state-of-the-art performance by up to 3.4 and 2.6 percent better than existing baselines, and by up to 3.4 and 3.8 times smaller fairness gaps between heterogeneous clients. FairFedTransNet is a strong privacy-sensitive rare disease diagnosis solution in real-world healthcare networks by integrating predictive accuracy, fairness, and interpretability in a scalable federation framework. The key FairFedTransNet innovations are a dual-stream transformer to structured and temporal EHR modelling, multimodal attentionguided fusion, fairness-aware adaptive aggregation, MOEA-based optimization, and adaptive differential privacy integration, which allow collaboratively fair, accurate, and privacy-preserving rare disease diagnosis.

Fields of Study & Subjects

Privacy-Preserving Technologies in Data Machine Learning in Healthcare Artificial Intelligence in Healthcare and Education

Publication details

Journal / source
International Conference on Communication, Computing and Electronics Systems (ICCCES)
Pages
1–9
Publication date
21 January 2026
Language
EN