[4Xin1] Poster session 2

[4Xin1-31] Prediction of neocortical gamma oscillations by pen accelerometers

〇Tianyi Zhang¹, Yu Sato¹, Keiji Tatani², Takamasa Fukuda³, Yuji Ikegaya^1,4 (1.The University of Tokyo, 2.Stolia Co., Ltd, 3.Mitsubishi Pencil Co., Ltd., 4.Center for Information and Neural Networks)

Keywords:Concentration Prediction, LSTM Networks, Pen Acceleration

The purpose of this study is to predict "concentration" when doing handwriting tasks based on "pen movement". Eight university students and staffs were asked to perform an 80-minute Arabic transcribing task, and their electroencephalograms (EEGs) were recorded. Seven of the eight experimenters' data were trained in an LSTM (long short-term memory) network to construct a predictive model of concentration and the remaining experimenter's data was used to test the model accuracy. The gamma/delta ratio, as an indicator of concentration, was predicted from the "pen movements" of the experimenter. As a result, a sensitivity and specificity of more than 80% were achieved to identify the time period of concentration. This result proposes a new method to estimate brain states from everyday tools without measuring EEG and may be applied to various applications such as education and work in the future.