使用重疊球體正向模型以及階層搜尋光束構成進行腦電波訊號源定位

Abstract

腦，是人體當中最重要且複雜的器官，對於腦功能的探索，是目前最熱門的研究課題之一。腦電波儀是常被用來進行腦功能研究的工具之一，因為它具有相當高的取樣頻率以及低廉的價位。在本篇論文中，我們將會提出一個準確且快速的腦電波訊號源定位方法，包括了正向模型的計算（針對已知的訊號源計算腦電波），以及如何去解決反向問題（利用腦電波來進行訊號源定位）。 在所提出的正向模型中，我們針對每個腦電波儀感應器找出適合的重疊球體，以增加正向模型的準確度。並且，重疊球體只需要使用多層球體結構即可很快速的計算出來，而不需要去計算相當耗時的邊界元素模型。基於所提出的正向模型，我們使用階層搜尋光束構成來解決反向問題。經由最大化腦電波的活動狀態和控制狀態的強度對比，可以大幅地增加訊號源定位結果的準確度。並且，在腦空間中搜尋訊號源時我們使用階層搜尋而非全域搜尋，亦能有效地降低解決反向問題的時間。 根據假體實驗以及視覺刺激實驗結果，確認了我們所提出的方法之準確性以及便利性。我們所提出的訊號源定位法可以有效地被使用在無需磁振造影資料的情形，例如進行基礎腦科學的研究，或者腦機介面的開發。

Brain is the most important and complicated apparatus of human beings. EEG has been widely applied in functional brain studies due to its high temporal resolution and low cost. In this work, we focus on the development of an accurate and efficient EEG forward model as well as the inverse solution for neuronal source estimation from the EEG measurements.

Our forward model successfully gains its accuracy by fitting an overlapping sphere for each EEG sensor. The computation of the overlapping sphere requires only the multi-shell geometry, instead of boundary element method, thus the proposed forward model is easy to compute. Based on the proposed forward model, the beamforming technique is applied to calculate the distributed sources in the brain space. We maximize the power contrast between active state and control state of EEG recorded data to improve the accuracy of inverse solution. Hierarchical search in the solution space is applied to save the amount of computation by searching grid point level by level instead of searching the whole brain space.

According to our experiments using phantom data and visual-evoked potential data, the proposed forward model and inverse solution can efficiently and accurately estimate the source of brain activation. A quick and reliable source localization technique for EEG is successfully developed which can be applied on applications when MRI is not available, such as fundamental brain research and brain-computer interface.