Federated Learning with Lion Optimization for Clinically Interpretable Multi-Class Thyroid Disorder Diagnosis.

Abstract: Artificial intelligence has increasingly been adopted in healthcare for disease diagnosis; however, concerns regarding data privacy, heterogeneity of clinical data, and lack of interpretability remain significant challenges. Federated learning has emerged as a promising paradigm that enables collaborative model training across distributed healthcare institutions without sharing sensitive patient data [1], [9]. In this study, a Lion Optimization-based Federated Learning framework (LbFL-TDP) is proposed for multi-class thyroid disorder diagnosis. The proposed model leverages distributed clinical datasets containing key endocrine biomarkers such as TSH, T3, and T4, ensuring both diagnostic relevance and privacy preservation. To address the limitations of traditional aggregation methods under non-IID data conditions, the Lion Optimization Algorithm is integrated to enhance global model convergence and predictive performance. The model is evaluated using cross-validation and achieves an accuracy of approximately 92% with an AUC of 0.95, outperforming conventional machine learning approaches. Furthermore, feature importance analysis demonstrates alignment with established endocrinological knowledge, improving model interpretability. The findings suggest that the proposed framework provides a robust, privacy-preserving, and clinically meaningful solution for thyroid disorder classification, highlighting the potential of federated learning in real-world healthcare applications [3], [7].

Keywords: Thyroid Disorder Classification, Federated Learning, Lion Optimization Algorithm, Non-IID Data, Distributed Machine Learning.