Federated Adaptive Personalized Optimization (Fed-APO): A Meta-Learning Approach to Enhancing Healthcare for Non-IID Multi-Healthcare Data

Federated Learning (FL) becomes a progressive solution enabling confidential model collaboration training in healthcare settings. FL algorithms encounter significant limitations from medical data structures that maintain non-independent distributed characteristics thus leading to inefficient models while slowing down training time. The researchers introduce Fed-APO as their new framework that uses meta-learning aggregation techniques with individualized training protocols to improve the performance of FL systems working with diverse healthcare data. Fed-APO carries out model updates through individualized characteristics of clients to optimize global-local collaboration by implementing learning rate adaptation and weight adaptation methods. The proposed method receives evaluation from experimental tests performed on medical datasets featuring different attributes for heart disease prognosis and general health monitoring along with cancer identification. The research outcomes demonstrate that Fed-APO provides superior performance compared to standard FL techniques Fed-Avg and Fed-NOVA through its advanced accuracy levels and F1-score capabilities and decreased communication requirements. Fed-APO elevates accuracy by 5.55% more than Fed-Avg and 4.26% better than Fed-NOVA when communication rounds decrease by 16.7%. The Fed-APO system achieves exceptional performance scalability in medical applications through its multidimensional organization model training method which safeguards patient privacy at all times