憂鬱症腦波分析

Analysis of Depression EEG

鄭幃騰(Wei-Teng Zheng)

指導教授 ： 劉益宏

中原大學/工學院/機械工程研究所/碩士(2014年)

https://doi.org/10.6840/cycu201400893

摘要

聯合國世界衛生組織(WHO)研究指出，在2020年全世界有三大疾病需要重視，其中之一即為憂鬱症(Depression)，並估計憂鬱症人口約為總人口數的3%，而造成人類失能前十名的疾病，第一名也是憂鬱症；本論文提出了辨識憂鬱症的方法，在未來對於憂鬱症檢測上，能夠實質的協助醫師及病患。
本論文基於憂鬱症腦波(EEG)辨識而設計了一套實驗流程，實驗分為放鬆階段與情緒誘發階段，其中放鬆階段在實驗中會讓受測者觀看灰屏十字的畫面，請受測者心情保持放鬆；而情緒誘發階段在實驗中使用了54張情緒圖片提供受測者進行實驗，其情緒圖片皆源自於國際情緒圖片系統(International Affective Picture System, IAPS)的High Valence–High Arousal(HVHA)圖庫，此類圖庫的圖片皆屬於正向情緒。實驗將受測者分為憂鬱症患者與健康者兩大類，每一類各有12位受測者，兩類性別相互對應且年齡也對應在正負差2歲以內，共計24位受測者。每位受測者都有54筆觀看灰屏十字與情緒圖片的腦波訊號，利用此數據當作機器學習分類的依據並進行投票，不過偶數筆資料進行投票可能會造成同票的情況，故本研究分析時改用每位受測者所擷取到腦波訊號之前53筆資料，如果53筆資料中超過一半判別為憂鬱症患者，則判定此受測者目前為憂鬱狀態，反之53筆資料中超過一半判別為健康者，則判定此受測者目前為健康狀態。
本實驗特徵抽取方法包括頻帶功率(Band Power, BP)、共同空間模式(Common Spatial Pattern, CSP) 和BP、CSP搭配主成份分析(Principal Component Analysis, PCA)特徵抽取串聯方法，而分類器有K個最近鄰居法(K-nearest neighbor, K-NN)、支持向量機器(Support Vector Machine, SVM)。實驗結果顯示特徵抽取使用CSP搭配PCA，利用SVM分類器能夠在頂葉(Parietal)上達到100% 的分類效果；此外也發現放鬆階段的分類率整體而言比情緒誘發階段來的好，因此情緒誘發的機制將可能不再需要；最後對於實驗的試驗(Trial)次數進行探討，發現試驗次數在15~27次就有相當程度的分類效果。綜合以上所得到的結果在未來將有助於開發即時檢測憂鬱症系統。

關鍵字

憂鬱症、腦波訊號、特徵抽取、主成分分析、支持向量機

並列摘要

According to World Health Organization(WHO), by year 2020 three diseases will need extra attention worldwide, one of them is Depression, estimated about 3% of the human population are suffering from Depression, and Depression is the number one reason on the top ten list of diseases that result in disability. A technique to distinguish Depression is proposed in this thesis, providing essential support in Depression detection for medical personnel and patients in the future.
In this thesis, an experimental procedure was designed based on Depression EEG recognition, the experiment consists of two states, resting state and emotion-induction state, in resting state a gray background with a cross in the middle (neutral image) is presented to the participant on the screen, and the participant is asked to rest and stay calm. In emotion-induction state, 54 pictures from International Affective Picture System’s (IAPS) High Valence – High Arousal (HVHA) database are used, all categorized as positive emotion. The experiment sorts participants into two categories, with and without Depression, each category consists of 12 participants, with corresponding gender and age with deviation within 2 years, a total of 24 participants took part in the experiment. 54 EEG signals are collected from each participant when the neutral image is presented, these data are used for machine learning classification and conducting a vote, since voting with a total of even number data might results in a tie vote, in this thesis uses the first 53 data for analysis, if more than half of these data are classified as depressed, the participant will be sorted as in a depressed state, on the other hand, if more than half of these data are normal, the participant will be judged as in a healthy state.
In this thesis feature extraction methods including Band Power (BP), Common Spatial Pattern (CSP), and a combination of BP, CSP and Principal Component Analysis (PCA) feature extraction method are adopted, K-nearest neighbor (K-NN) and Support Vector Machine (SVM) are the classifiers used. Experimental results show that using CSP along side with PCA for feature extraction and SVM for classification can achieve 100% classification rate at Parietal area. Furthermore overall classification rate at resting state is higher than at emotion-induction state, therefore emotion-induction state might not be needed in the future. An investigation on the number of trials in the experiment reveals that when the number of trials conducted in the experiment is between 15 to 27 times could give a certain level of classification rate. Summing all the results could provide essential support for future real-time Depression detection system development.

並列關鍵字

SVM ； Depression ； EEG ； Feature Extraction ； PCA

參考文獻

[6] 王治漢, 情緒腦波誘發範例建立及辨識,中原大學機械工程研究所碩士學位論文, 2012.

[27] 高永樺, 情緒腦波分析及腦機介面之開發,中原大學機械工程研究所碩士學位論文, 2013.

[28] 黃俊偉, 非同步腦機介面方法比較及即時腦控開關之發展,中原大學機械工程研究所碩士學位論文, 2013.

[2] R. Caton, “The electrical currents of the brain,” The Journal of Nervous and Mental Disease, vol. 2, no. 4, pp. 610, 1875.

[3] H. Berger, “Über das Elektrenkephalogramm des Menschen,” European Archives of Psychiatry and Clinical, vol. 87, no. 1, pp. 527–570, 1929.