We have been developing a cognitive brain-machine interface (BMI) system, "Neurocommunicator" to support augmentative and alternative communication (AAC) for people with severe motor disabilities. The system can predict a user´s decision about a message to be expressed by his/her CG avatar, which is based on a real-time analysis of electroencephalography (EEG) data, especially event-related potentials (ERPs) such as P300. The accuracy of the model, which was generated by linear discrimination analysis (LDA), at single predictions of the target message (a pictogram) was generally over 95%, corresponding to about 32 bits per minute for normal subjects.
So far, we have focused on improvements of the hardware for recording of EEG data. In the first major upgrade, the custom–made plastic headgear (ver. 2.0) was introduced instead of a close-fitting cloth cap (ver. 1.0); the plastic headgear improved both decoding accuracy and usability in normal subjects. The decoding accuracy in the patients, however, did not reach the level of normal subjects mainly because there were full of electrical noise in their houses, which made it difficult to measure EEG in good condition.In this study, as the second major upgrade, we have introduced a new headgear (ver. 3.0), which was covered by an electromagnetic shield. The new system reduced the noise and improved the decoding accuracy even in the patients, suggesting it as a practical solution.