This paper proposes a Secured Medical Homomorphic Encryption Algorithm (SMHEA) to ensure privacy in medical data using Federated Learning (FL) in an Internet of Things (IoT)-based healthcare system. FL allows remote devices to train a learning system without sharing their data, enhancing privacy by transmitting only localized model variables. However, there is a risk of sensitive data exposure through analysis of regional learning networks. SMHEA integrates homomorphic encryption with secure computing to protect medical data privacy. Cryptographic techniques like masking and homomorphic cryptography are used to secure local modeling, preventing adversaries from deducing confidential health information. The primary determinant for assessing regional modeling contributions is the quality of the databases rather than their size. A dropout-tolerant approach is suggested, allowing FL to continue as long as the total number of online customers remains above a certain level. Security evaluation shows that the proposed approach effectively ensures data privacy. Theoretical analysis is conducted on accuracy, computation time, and communication error. An example of clinical application is the categorization of skin lesions using training photos from the HAM10000 medical database. Experimental results demonstrate that the proposed system achieves favorable performance and privacy preservation outcomes compared to current methods. The SMHEA framework includes a model aggregating server and dispersed consumers. The server gathers anonymized local models from consumers and executes actions to finalize safe aggregation. Clients are medical institutes with substantial unprocessed health information. They train regional models using local medical databases and send masked regional models and secure variables to the server. The server and clients collaborate to construct the ideal universal architecture. A third-party Trust Agency (TA) is included to bolster privacy protection. The TA initializes various security variables, including public and private credentials. The server and customers are assumed to be trustworthy but inquisitive, willing to follow the protocol but intending to secretly gather confidential details. The system ensures privacy by limiting access to masked modeling and secure variables. The SMHEA uses a decentralized homomorphic encryption approach, dividing private keys into partial confidential keys and distributing them to multiple dispersed computers. The system generates public-private credentials, encrypts and decrypts data using these keys, and performs fractional decryption and aggregation. The security study shows that the masking technique effectively safeguards actual inputs. The system determines if colluding entities can decrypt information and recover localized modeling. The server must first re-establish the value of 2^{Q_{x}^{p}} to decrypt information quality. The private key is securely maintained, and the server must solve the distinct logarithm issue to determine the private key. If the server cannot recover the private key, it conspires with other customers to seek their aid in rebuilding the secret credential. The system ensures the safeguarding of localized modeling by preventing the reconstruction of the private key. The system evaluation shows that the SMHEA achieves higher accuracy, shorter computation time, and lower error rates compared to other FL approaches. The SMHEA's effectiveness is attributed to the integration of HEThis paper proposes a Secured Medical Homomorphic Encryption Algorithm (SMHEA) to ensure privacy in medical data using Federated Learning (FL) in an Internet of Things (IoT)-based healthcare system. FL allows remote devices to train a learning system without sharing their data, enhancing privacy by transmitting only localized model variables. However, there is a risk of sensitive data exposure through analysis of regional learning networks. SMHEA integrates homomorphic encryption with secure computing to protect medical data privacy. Cryptographic techniques like masking and homomorphic cryptography are used to secure local modeling, preventing adversaries from deducing confidential health information. The primary determinant for assessing regional modeling contributions is the quality of the databases rather than their size. A dropout-tolerant approach is suggested, allowing FL to continue as long as the total number of online customers remains above a certain level. Security evaluation shows that the proposed approach effectively ensures data privacy. Theoretical analysis is conducted on accuracy, computation time, and communication error. An example of clinical application is the categorization of skin lesions using training photos from the HAM10000 medical database. Experimental results demonstrate that the proposed system achieves favorable performance and privacy preservation outcomes compared to current methods. The SMHEA framework includes a model aggregating server and dispersed consumers. The server gathers anonymized local models from consumers and executes actions to finalize safe aggregation. Clients are medical institutes with substantial unprocessed health information. They train regional models using local medical databases and send masked regional models and secure variables to the server. The server and clients collaborate to construct the ideal universal architecture. A third-party Trust Agency (TA) is included to bolster privacy protection. The TA initializes various security variables, including public and private credentials. The server and customers are assumed to be trustworthy but inquisitive, willing to follow the protocol but intending to secretly gather confidential details. The system ensures privacy by limiting access to masked modeling and secure variables. The SMHEA uses a decentralized homomorphic encryption approach, dividing private keys into partial confidential keys and distributing them to multiple dispersed computers. The system generates public-private credentials, encrypts and decrypts data using these keys, and performs fractional decryption and aggregation. The security study shows that the masking technique effectively safeguards actual inputs. The system determines if colluding entities can decrypt information and recover localized modeling. The server must first re-establish the value of 2^{Q_{x}^{p}} to decrypt information quality. The private key is securely maintained, and the server must solve the distinct logarithm issue to determine the private key. If the server cannot recover the private key, it conspires with other customers to seek their aid in rebuilding the secret credential. The system ensures the safeguarding of localized modeling by preventing the reconstruction of the private key. The system evaluation shows that the SMHEA achieves higher accuracy, shorter computation time, and lower error rates compared to other FL approaches. The SMHEA's effectiveness is attributed to the integration of HE