以深度學習為基礎之腦電波特徵選取與分類分析

Deep Learning-Based EEG Feature Selection and Classification Analysis

林子堯(Tzu-Yao Lin)

指導教授 ： 歐陽振森

義守大學/智慧科技學院/資訊工程學系/碩士(2021年)

摘要

腦電波訊號通常為多頻道時間序列，其各頻道對於分類或迴歸分析目標的重要性不一。為了減少資料多餘性並增加計算效率性，如何挑選出重要的頻道已成為腦電波分析領域中相當重要且急迫的研究議題。本研究針對腦電波訊號頻道挑選問題提出基於腦電波連結性的頻道挑選方法，再透過卷積神經網路分類模型驗證其對於腦電波分類問題的優越性。首先，計算訓練資料集中腦電波訊號各頻道的連結性總和。接著，挑選出在分類類別之間較具有統計顯著差異的數個頻道，再對訓練資料進行頻道縮減。最後，訓練一個卷積神經網路分類模型，再以測試資料集中縮減頻道後的腦電波訊號測試已訓練模型的分類表現。上述模型分類表現評估乃是採用重複交叉驗證策略。此外，本研究亦比較了多種連結性計算子與其他腦電波頻道選擇方法。實驗結果顯示相較於其他方法，本方法在大部分的選擇頻道數條件下，具有較佳的分類表現。

關鍵字

腦電波 ； 分類分析 ； 腦電波頻道選擇 ； 特徵選擇 ； 連結性 ； 卷積神經網路

並列摘要

Electroencephalogram (EEG) signals are usually multi-channel time series and the importance of channels for the goal of classification or regression analysis are different. To reduce the data redundance and increase the computation efficiency, how to select important channels has been an important and urgent research issues in EEG analysis. In this study, we propose an EEG connectivity-based channel selection for the EEG channel selection problem and verifies its superiority for EEG classification problems through convolutional neural networks. First, the connectivity sum of each channel of EEG signals in the training dataset is calculated. Then, serval channels with higher statistically significant difference among classification classes are selected, and then channel reduction on the training dataset is performed. Finally, a classification model based on convolutional neural network is trained, and then the classification performance of trained model is tested with channel-reduced EEG signals in the test dataset. The above evaluation of model’s classification performance is based on a repeated cross validation strategy. Moreover, several connectivity measures and the other EEG channel selection approach are also compared in this study. Experimental results have shown that compared with the other approach, our approach possesses the better classification performance in most cases of selected channel sizes.

並列關鍵字

electroencephalogram ； classification analysis ； channel selection ； feature selection ； connectivity ； convolutional neural network

參考文獻

[1] L. R.Hochberg et al., “Neuronal ensemble control of prosthetic devices by a human with tetraplegia,” Nature, vol. 442, no. 7099, pp. 164–171, 2006.[2] T. J.Oxley et al., “Motor neuroprosthesis implanted with neurointerventional surgery improves capacity for activities of daily living tasks in severe paralysis: First in-human experience,” J. Neurointerv. Surg., vol. 13, no. 2, pp. 102–108, 2021.

[3] J. S.Kumar and P.Bhuvaneswari, “Analysis of electroencephalography (EEG) signals and its categorization - A study,” Procedia Eng., vol. 38, pp. 2525–2536, 2012.

[4] A.Antoniades et al., “Detection of interictal discharges with convolutional neural networks using discrete ordered multichannel intracranial EEG,” IEEE Trans. Neural Syst. Rehabil. Eng., vol. 25, no. 12, pp. 2285–2294, 2017.

[5] A. S.Al-Fahoum and A. A.Al-Fraihat, “Methods of EEG Signal Features Extraction Using Linear Analysis in Frequency and Time-Frequency Domains,” ISRN Neurosci., vol. 2014, pp. 1–7, 2014.

[6] A. M.Bastos and J. M.Schoffelen, “A tutorial review of functional connectivity analysis methods and their interpretational pitfalls,” Front. Syst. Neurosci., vol. 9, no. JAN2016, pp. 1–23, 2016.