聴覚事象関連電位への神経デコーディングの適用 : 統計的識別手法の比較と脳波分析方法としての評価(<特集>脳機能計測と基礎心理学)

概要

論文の詳細を見る
Our aim was to predict the duration of a stimulus, which was a tone or a gap in a continuous tone, from auditory event-related brain potentials (neural decoding). We also aimed to compare performances across statistical classification methods for prediction. We found that the decoder performance was significantly higher than the chance level. The naive Bayes method with principal-component-analysis preprocessing (PCA+NB classification) and the support-vectormachine method (SVM classification) revealed a higher performance than the naive-Bayes method (NB classification) alone. Erroneous classifications were distributed in the neighborhood of the correct class. It is suggested that the SVM method has an advantage for its high performance without parameter optimization, while the PCA+NB method has an advantage for clarifying brain representation. Statistical classification is thus shown to be an effective method for analyzing an EEG and could be useful for investigating neural representation of sensory input.
2009-09-30