FedXAI-Health: Privacy-Protected Explainable Federated Deep Learning Model in Pre-emption of Early Diseases with Multimodal Healthcare Data

Abstract

Machine learning (ML) has significantly accelerated healthcare analytics, but its widespread clinical implementation is hindered by issues of data privacy, inter-institutional heterogeneity, and a lack of interpretability. To overcome these challenges, this paper proposes FedXAI-Health, a novel framework that combines federated learning (FL), multimodal deep learning, and explainable artificial intelligence (XAI) to predict diseases while preserving privacy. The proposed methodology enables multiple healthcare facilities to collaboratively train a global model without transferring raw patient information. A hybrid model combining CNNs for medical imaging and transformer-based models for EHRs is used. SHAP-based explainability is added to increase transparency. Tested on MIMIC-III and chest X-ray data, the model achieves an accuracy of 94.2% and an AUC of 0.96.