Federated learning (FL) is a distributed machine learning system that enables local data sharing for medical IoT devices and EHRs, streamlining healthcare processes and promoting user data privacy. However, FL faces privacy challenges due to model updates being shared between clients and servers, potentially breaching privacy requirements. Healthcare organizations need to address privacy issues in machine learning due to digitization of patient data. Federated learning allows medical institutions to train deep models collaboratively using homomorphic encryption. This study aims to improvise the erstwhile privacy-preserving systems with federated learning system for medical and healthcare data by implementing a realistic implementation of a preprint with a proposed encryption technique (MK-CKKS). The solution involves adjusting a basic “ring learning with error” method, forking a federated learning framework for Python, and incorporating a communication process with protocol buffers. Experimental assessments confirm the framework’s strong model performance despite these adjustments.