Federation learning (FL), a kind of decentralized machine learning, may secure client data from malicious actors. When examining client-uploaded parameters, including model-trained weights, sensitive data might be disclosed. Current methods are limited by data privacy issues and the time required to train and test with large datasets. We presented an enhanced federated learning-based privacy preservation technique for acute bladder inflammation detection. Like differential privacy (DP), this method augments clients’ disease characteristics with artificial noise before aggregation. We demonstrated that our method can address sounds and fulfil DP at different protection levels. Next, we computed the model’s loss and accuracy using the enhanced privacy preservation approach on the acute inflammatory dataset to evaluate its performance. When it comes to protecting model updates from malicious users, federated learning finds the perfect balance between privacy protection and model accuracy and loss using differential privacy techniques. The training accuracy will approach 99.85% and the loss of the FL models will be 0.0015% as the number of iterations increases. This proves that the model performs better when a larger privacy budget is allocated to a higher iteration. It is observed that the model may offer greater privacy protection without compromising performance by allocating a bigger privacy budget to higher iteration counts.