The Internet of Medical Things (IoMT) emerges as a crucial component of e-health technologies, interconnecting various medical devices to collect, exchange, and analyze health information. However, the security significance of IoMT becomes increasingly critical due to the inherent user privacy concerns associated with medical data. Moreover, traditional centralized learning models for IoMT pose a high susceptibility to security attacks. Additionally, global machine learning approaches fall short in convergence speed due to the heterogeneity of data and systems and cannot provide personalized medical advice for diagnostic purposes. In this paper, we propose an adaptable heterogeneous Blockchain-based Personalized Federated Learning approach with Model Similarity (BPFL-MS). We first introduce a two-layer network framework and then conduct an in-depth analysis of personalization issues, accuracy challenges, and computational efficiency problems related to integrating federated learning with blockchain technology. Finally, we perform evaluations that demonstrate how BPFL-MS outperforms baseline algorithms in terms of model accuracy and time delay.