[3Yin2] Interactive session 1

[3Yin2-38] Deep learning model for discriminating between eye-closed resting state and anesthetized state based on electrocorticogram (ECoG) of macaque monkeys

〇Meguru Kobayashi^1,2, Takafumi Soda^1,3, Hiroyuki Yamaguchi¹, Misako Komatsu⁴, Manabu Honda¹, Yuichi Yamashita¹ (1.National Center of Neurology and Psychiatry, 2.Tokyo University of Agriculture and Technology, 3.Tokyo Medical and Dental University, 4.RIKEN Center for Brain Science)

Keywords:Convolutional neural network, Spectrogram image, Anesthesia, Machine learning, Electroencephalogram

Electrocorticogram (ECoG) is a promising method for measuring brain functions. However, owing to the data characteristics of high-dimensional time series, conventional analysis methods may overlook useful information in ECoG. To address this issue, in the current study, we developed an ECoG analysis method using a convolutional neural network (CNN). The CNN was tasked with discriminating between the eye-closed resting state (ECRS) and anesthetized state (AS) from ECoG in macaque monkeys. As a result of training, the CNN successfully discriminated between ECRS and AS with generalization for inter-recording day (98.5%) and inter-subject (92.9%), but with slightly decreased accuracy for inter-anesthetic drug (86.4%). Subsequently, sensitivity map analysis by SmoothGrad demonstrated that the decline in performance may be due to the difference in sensitivities to the frequency ranges associated with each anesthetic. These results demonstrated that the CNN could extract useful information from ECoG and contribute to basic research and clinical applications.