Bridging the Interoperability Gap in Healthcare AI: Adaptive Federated Learning for Secure, Cross-Platform Data Harmonization

Background: The adoption of AI in healthcare is hindered by data silos and privacy constraints, with heterogeneous sources like EHRs, wearables, and IoT devices operating in isolated ecosystems.Traditional centralized AI models fail to address interoperability while complying with regulations like HIPAA and GDPR.Federated learning (FL) has emerged as a promising decentralized alternative, but its adaptation for cross-platform healthcare data harmonization remains underexplored.Objective: This study proposes an adaptive FL framework to bridge interoperability gaps in healthcare AI, enabling secure collaboration across disparate data sources without centralized data pooling.We aim to (1) design a privacy-preserving FL architecture for heterogeneous medical data, (2) validate its interoperability using real-world EHR and wearable datasets, and (3) quantify compliance with regulatory requirements. Methods: Our framework integrates Federated Multi-Task Learning (FMTL) to handle data heterogeneity, inspired by [1]'s work on cross-institutional EHR analysis, and differential privacy (DP)enhanced aggregation, extending [2]'s approach forIoT-health data.We evaluate the system using two public datasets: (i) MIMIC-III (EHRs) and (ii) PPG-DaLiA (wearable photoplethysmography), simulating a multi-device environment.Interoperability is tested via schema mapping tools (e.g., FHIR standards), while privacy guarantees are audited using GDPR-specific metrics (e.g., data minimization, user consent logs).Results: The framework achieved 88.7% harmonization accuracy (vs.72.3% in centralized baselines) across EHR-wearable data fields, reducing semantic heterogeneity by 41% through adaptive FMTL.DP noise injection (< =1.5) maintained model utility (AUC drop < 3.2%) while ensuring compliance with HIPAA's de-identification criteria.Comparative analysis showed 23% faster convergence than conventional FL, attributed to dynamic client selection for heterogeneous devices [1].Regulatory audits confirmed adherence to GDPR's Article 35 (DPIA) and HIPAA's Safe Harbor rule.Conclusion: Adaptive FL can effectively harmonize fragmented healthcare data while preserving