The growing interest in ensuring patient privacy has led hospitals to increasingly restrict access to medical records and exams data. On the one hand, it has increasingly guaranteed security, on the other hand, it has led them to become silos of private data, making it difficult or even preventing the training of machine learning models to aid diagnoses. Seeking to reconcile the best of these two scenarios, swarm learning came as a framework that has enabled model training without the need for data sharing, unifying the distributed training of Federated Learning with the security of Blockchain. This paper introduces the structure and functioning of the Healthchain project, a swarm network that allows the training of multiple machine learning models and the sharing of results between different hospitals through a bidimensional architecture based on blockchain. A Proof of Concept (PoC) was developed to validate its operation. In the experiments performed, the network was used for transfer learning, sharing models and training, and for carrying out FedAVG filtered by nodes, a federated learning aggregation method that allows selecting the training that will be used for the calculation, avoiding the use of node training with low data quality. The paper also shows the evaluation of growth rates of the size of training blocks according to the data used.