Neural network recognition of electro-encephalogram patterns preceding voluntary movements

Abstract

In the consideration of the human interface of a machine or computer, it is important to understand the limitations on the ultimate interface. One possible ultimate interface would be an input interface based on the electroencephalogram (EEG) pattern recognition. This approach was used in the past to detect the evoked response accompanying visual/aural stimulation of human subjects from the EEG. This study focuses on the possible information input to a computer or machine based on the potential distribution pattern on the human skull. The readiness potential detected from the skull immediately before voluntary movement, such as the hand motion or voluntary utterance, is recognized by a neural network. The result of such an approach is reported in this paper. For this purpose, experiments should be made to detect the intention of motion, in which the subject only “thinks” of moving. As a first step toward the goal, however, the actual motion is executed in this study, and the recognition of the immediately preceding potential pattern is attempted. A recognition rate of 53% is obtained for the average readiness potential resulting from the utterance of five vowels. A recognition rate of 96% is obtained for the average readiness potential preceding the operation of a joystick in four directions. In experiments based on the principle of an actual EEG input device, recognition of the potential distribution on the skull preceding a single trial voluntary movement was attempted. A recognition rate of 90% was obtained for the intention to utter one of two vowels. A recognition rate of approximately 60% was obtained in recognition of the intention to move the joystick in one of three directions. These were all recognition rates for unknown test patterns, and the results suggest the possibility of a future EEG interface based on recognition of the potential distribution pattern on the skull. © 1997 Scripta Technica, Inc. Electron Comm Jpn Pt 3, 80(7): 65–73, 1997