探討認知狀態轉變下大腦神經網路建構模型與其應用研究

Abstract

疲勞或精神不濟的狀態下勉強駕駛是一種危險自身與公共安全之行為，不論是睡 眠不足、長途駕駛、深夜時段開車、服用嗜睡藥物、單調公路駕駛與睡眠障礙疾病等 都有可能引起瞌睡與疲勞駕駛，且是一般健康者都有可能面臨到的危險，有鑑於此， 本計畫將探討認知狀態轉變下大腦神經網路建構模型與其應用研究，提出具體三年規 劃，從基礎研究著手進行至應用研究，第一年開發特徵選取或選取基礎實驗，找出與 瞌睡最為相關腦區，與過去相關研究呼應。第二年探討清醒至瞌睡生理變化之大腦神 經網絡，第三年將發現的腦區特徵實現於腦機介面，展示即時估算演算法。本研究將 以虛擬動態場景模擬真實駕車環境，主動擷取駕駛人的腦電波訊號，以時頻分析做為 主要基礎分析工具，並使用Granger Causality Mapping 與Dynamic Causal Modeling 分 析駕駛者大腦各區位之功能的連結性與因果關係，進而萃取在神經網路中調控駕駛行 為之相關頻率，研究其在清醒期、過度期與瞌睡期之神經網路變化，以上述之發現作 為開發瞌睡辨識器系統的重要特徵，深入探究不同機器學習與圖形識別技術在此資料 上的適合度、效能與正確性，作為未來發展腦機系統中的重要核心技術之參考。

Drowsy or fatigue driving could endanger personal and public safety. To fall into a drowsy or fatigue state might be possibly consequent on sleep deficit, long-term driving, midnight driving, monotonous driving, taking sleeping drugs, or sleep disorders, etc. This is a common occurrence that anyone, even for a health person, could confront with. For this serious reason, the development of a countermeasure to fight drowsy driving is a practical and valuable research. Therefore, we propose a three-year planning to implement the brain computer interface with real-time cognitive state monitoring algorithm. In the first year, we will develop an automatic feature selection/extraction method and compare with the major findings with the drowsiness related studies. In the second, we will study how the brain network works from alert to drowsiness via Granger Causality and Dynamic Causal modeling. In the last year, we will implement the major brain features of drowsiness into the mobile brain computer interface. We can also develop a real-time cognitive state monitoring algorithm. A realistic driving environment with a motion platform is used to perform driving task. The subject as a driver has to wear an EEG cap during driving. The time frequency signal analysis, Granger causality mapping, and dynamic causal modeling are used as fundamental tools to analyze the relation and causality between brain regions. In the further step, we will identify linear or nonlinear interactions in brain network and extract the important frequency and coupling parameters which govern the network in different cognitive states. Finally, all these informative factors are taken into account as importance features to investigate the performance by applying various machine learning methods. We expect this comprehensive study can provide the important knowledge for studying drowsy driving and designing BCI equipment in the future.