Design of a BMI for tetraplegic patients

Tetraplegia is a paralysis of the four limbs, which leaves attained people without their full autonomy and freedom. Indeed, tetraplegia not only affects the movement capabilities of the patients, but also their ability to communicate with others. Needless to say that tetraplegic patients’ everyday life is not easy at all and that they constantly suffer. In this context, a very interesting and exciting prospect is the field of Brain-Machine Interfaces (BMIs). BMIs, also called Brain-Computer Interfaces (BCIs) or neuroprosthetics, can be defined as systems that use brain signals as inputs to produce a control signal as output. This output can be used in various kinds of applications: to control a wheelchair, to move a mouse cursor, to control a voice synthesizer and communicate with other people. With a BMI, tetraplegic patients could move or communicate in an autonomous way, just by thinking about the action they would like to perform. Obviously, BMIs have the potential to change the life of tetraplegic patients, and help them retrieve a kind of autonomy. The primary objective of this work has been to imagine and design an experimentation system, in collaboration with Synergia Medical. Devoted to the development of BMIs, and more particularly to EEG-based BMIs, this system is flexible, programmable and cheap. It was built around a FPGA, which is the central component of the electronic board. The board includes also ADCs to sample and digitize acquired EEG data. A microcontroller unit is available as well. To be used in a wide range of applications, many different types of interfaces were added to the design, such as USB, RF or MicroSD. In addition, to position correctly the EEG electrodes on users’ head, a helmet has been printed in 3D. Every feature of the system was imagined to ease the development of BMIs. To demonstrate that it is working properly, a small application has also been developed. Its goal is quite simple: when a user moves his right/left arm, the experimentation system lights the rightmost/leftmost LED of the board in response. The decoding of the user’s movements is based on Support Vector Machine (SVM) classifiers, which classify in real-time a feature vector. The features are autoregressive (AR) coefficients computed on small intervals of time for three acquisition channels. The performances are not bad nor extraordinary, but prove that the developed experimentation system is well suited for the design and testing of real-time BMIs. There is room for improvements, either to obtain better results with the already available application or to develop new and more complex ones. The world of BMIs is vast and a lot of unresolved questions need answers. The system developed in this work is definitely appropriate to address these topics.