Early disease detection is essential for effective treatment and improving patient outcomes. Traditional centralised machine learning (ML) methods for illness diagnosis, however, restrict access to geographically scattered healthcare facilities and create privacy problems. Deep learning models could make it easier to diagnose illnesses accurately and quickly. By constructing models in a distributed fashion and guaranteeing patient privacy, federated learning can be useful in solving these issues. In order to reduce communication overheads and resource constraints, the paper provides information on pre-processing methods, data sets, and artificial intelligence (AI) models utilised in the federated learning (FL) framework (such as deep learning convolutional neural networks). The results show that AI-based FL has the ability to diagnose diseases in a decentralised manner with excellent accuracy. Spatial inconsistencies were addressed by data preparation approaches, and generalisation was enhanced through data augmentation. The study also investigated blockchain integration with the FL system to guarantee model convergence and address data security concerns. By the end of the sixth training round, the FL model’s accuracy in identifying HCM had increased to between 85 and 95%. The efficacy of data collection and iterative training was demonstrated by the model’s steady performance improvement over rounds. According to the study’s findings, federated learning is a viable option for multi-centre cardiovascular diagnosis that maintains high accuracy while protecting patient privacy. These results provide credence to FL’s potential as a scalable, privacy-preserving healthcare solution for a range of healthcare facilities.