Meta-XPFL: An Explainable and Personalized Federated Meta-Learning Framework for Privacy-Aware IoMT

In the Internet of Medical Things (IoMT), specifically in the field of medical image classification—particularly for skin cancer detection—traditional methods face challenges related to data privacy, heterogeneity, and the need for personalization across institutions. This research proposes a personalized federated learning (PFL) framework Meta-XPFL that addresses these challenges through a decentralized approach, allowing institutions to collaboratively train models without sharing raw data. The framework integrates meta-learning for adaptability, and self-supervised learning to leverage unlabeled data and secure multiparty computation (SMPC). Adversarial training improves model robustness, while attention mechanisms enhance the focus on relevant image features. The use of explainable AI techniques ensures interpretability, which is crucial in clinical settings. To validate the proposed framework, experiments were conducted on the HAM10000 dataset for skin cancer classification, demonstrating significant improvements in model accuracy, privacy preservation, and robustness against adversarial attacks compared to traditional methods. The results indicate that the framework not only enhances scalability and diagnostic accuracy but also offers a privacy-preserving solution that can be extended to various types of medical images, making it adaptable for broader applications in IoMT.