With the rapid development of healthcare-based cyber physical systems (CPSs), more and more healthcare data is collected from clinical institutions or hospitals. Due to the private and fragmented nature, healthcare data is quite suitable to be processed by federated learning (FL) paradigm, where a shared global model is aggregated by a central server while keeping the sensitive healthcare data in local hospitals. However, there are two practical issues: (1) the centralized FL server may not honestly aggregate the final model, and (2) the FL participants lack incentive to contribute their efforts. In this study, we propose a blockchain-empowered FL framework for healthcare-based CPSs. A distributed ledger is maintained by a task agreement committee which is composed by the representators of the hospitals who execute FL tasks. A secure FL task model training-based consensus process is proposed to generate consistent blocks. Furthermore, a contribution point-based incentive mechanism is designed to fairly reward FL participators for contributing their local data. We evaluate the proposed system base on real healthcare data and the numerical results demonstrate its effectiveness in achieving FL model aggregation truthfulness and efficiency in providing incentives for FL participants.