Developing A Standard Computing Platform For Primate Brain-computer Interfaces

The brain-computer interface (BCI) provides a direct communication and control channel between the brain and external devices. The study of brain-computer interface requires complex cross research backgrounds, including computer science, biomedical engineering, and neuroscience. With the deepening of study, the experimental process of brain science becomes more and more complicate. A common procedure of BCI research contains experimental operation, the training of experiment paradigms, the design of algorithms, the implementation of hardware, and the construction of software platform, which are usually accomplished by different persons. It is quite difficult for them to communicate with each other because of the different background requirements in different stages in BCI. Therefore, it is necessary for us to standardize the brain-computer interfaces. This thesis focuses on the computing part in the BCI framework. Specifically, the scalability of the data algorithm, the configurability of process, and the support of big data computing are involved. And based on the case of a primate motor BCI system, we aim at developing a standard computing platform for the practitioners in this field.First, this thesis describes the specific processes of BCI experiments through a macaque motion decoding system. It includes the design of experimental paradigm, the development of hardware and software platform, and the data processing. Especially, this thesis mainly introduces the functional design of the paradigm module and the decoding module of the system. As for the data related to the decoding experiments, the data collection and the experimental paradigms are described in detail. And we have developed the unified format of experimental data to standardize the BCI technology from a data perspective.Then, this thesis introduces the design of BCI standard computing platform. As for the scalability and the configurability, we introduce the technology of scientific workflow. All configurations and interactions related in the decoding processes are encapsulated as simple visual nodes, including the datasets, the pre-processing methods, the decoding algorithms, and the display of experimental results. The researchers can configure the experimental procedure through graphical operation mode. The extendable interfaces of data and algorithms are also provided, which makes it easy to extend the computing platform. And the dependency detection mechanism in our workflow engine can effectively ensure the accuracy of the decoding process. In terms of the performance of the big-data computing, the design of the system architecture based on the hadoop distributed computing framework is also introduced in this thesis.Finally, this thesis describes the implementation details of the computing platform from such aspects as the system architecture, the communication protocols, and the function modules. The macaque motion data and the GRNN decoding algorithm are selected to test the performances of system, including accuracy, efficiency, and stability. It is proved that our BCI standard computing platform is scalable, configurable and it supports massive data parallel computation.