癲癇(Epilepsy)為一中樞系統的慢性疾病，其臨床特徵為腦部異常放電，並可引起抽搐或是失神等現象，且在無誘因情況下，有反覆發作的情形。臨床上，腦電圖(Electroencephalography, EEG)為檢測癲癇最廣泛的工具，常用來觀察癲癇發作並診斷異常放電的病灶。目前，醫生由腦電圖判斷是否有癲癇發作需要耗費大量的時間與心力。是以，為使癲癇腦電圖判斷更為方便且快速，發展一個自動化的分析方法，有其必要性。為此，本研究提出了一基於卷積神經網路(convolutional neural network, CNN)模型之分析方法，透過輸入多通道腦電圖至模型中，我們欲偵測患者是否進入發作期(ictal)。一開始，我們將所測量之癲癇腦電圖數據依「原始數據」、「濾波後數據」及「頻率是否發生變化」分為兩大類：發作間期(interictal)與發作期。由於腦電波所記錄的訊號為非穩態(non-stationary)且每個人的數據皆具有個體的差異性，故在模型訓練以前，我們先計算多通道統計值與非線性特徵以保留通道的結構與防止資訊的遺失，隨後再透過我們提出的卷積神經網路模型進行癲癇偵測。本論文使用癲癇在顳葉起源的數據進行偵測，所提取的特徵為腦電圖中各通道的標準差及樣本熵，平均延遲時間為10.3秒，準確率為78%，靈敏度為74%，其效果較基於卷積神經網路提取時域訊號特徵要佳，且為了有更好的泛化性，從原本的數據加入了癲癇在大腦中不同地方起源的數據，特徵也從原本的2個，增加為9個，平均延遲時間為8.6秒，準確率為74%，靈敏度為72%。

Epilepsy is a chronic neurological disease of the central nervous system. It is caused by abnormal brain discharge, accompanied by symptoms such as convulsions or lapse of consciousness. In clinical practice, electroencephalography (EEG) is the most comprehensive tool used to observe epileptic seizures for diagnosis. At present, doctors need to spend much time and effort to determine when an epileptic seizure occurs, and an assistive system is required to release the burden. To realize such a system, a convolutional neural network model is presented in this study. By feeding multi-channel EEGs to the model, the system is able to determine if it is ictal or not. Note that multi-channel EEGs have more information to discriminate seizures as compared with single-channel ones since the temporal information for time series generation and the interrelations among the electrode in the recordings of an epileptic event can be extracted to enable valid detection. The study begins with the data of patients with temporal lobe epilepsy. The extracted features are standard deviation and sample entropy, and the results demonstrated that an average delay of 10.3 seconds with 78% accuracy and 74% sensitivity was attained. To better generalize the model, data of different types of focal onset epilepsy were included for model training, and the average delay time was 8.6 seconds, the accuracy rate was 74%, and the sensitivity was 72%.

摘要 i
Abstract ii
誌謝 iii
目錄 iv
圖目錄 viii
表目錄 ix
第一章 緒論 1
1.1癲癇 1
1.2研究動機 1
1.3文獻回顧 2
1.4研究架構 4
第二章 數據前處理 6
2.1 數據模型與假設 6
2.2 數據來源 6
2.3 數據預處理 6
2.3.1 數據切割 7
2.3.2電極連接方式 7
2.3.3 濾波 7
2.3.4 去除干擾波 8
第三章 小波轉換與特徵萃取 9
3.1 小波轉換 9
3.1.1 離散小波轉換 9
3.1.2 濾波器組(Filter bank) 10
3.2 特徵萃取 11
3.3 特徵矩陣(Feature matrix) 12
第四章 深度學習 13
4.1 神經網路的介紹 13
4.1.1 神經網路 13
4.1.2 活化函數(Activation function) 14
4.1.3 深度神經網路 15
4.2 神經網路的學習 15
4.2.1 損失函數 15
4.2.2 最佳化方法 16
4.2.3 過擬合(Over-fitting) 17
4.3 卷積神經網路 18
4.3.1 卷積層(Convolution layer) 18
4.3.2池化層(Pooling layer) 19
4.4 卷積神經網路的架構 20
4.5 預測後處理 21
第五章 實驗設計與結果討論 23
5.1 實驗規劃 23
5.1.1 癲癇判別 23
5.1.2 數據的挑選 24
5.1.3 泛化前的方法比較 24
5.1.4 泛化性模型 24
5.1.5 比較指標 25
5.2 實驗結果與討論 26
5.2.1 濾波與消除干擾波 26
5.2.2 與現有方法比較 28
5.2.3 增加模型泛化性的結果 29
第六章 總結與未來展望 31
6.1 總結 31
6.2 未來展望 31
參考資料 33
 

[1] A. Shoeb, H. Edwards, J. Connolly, B. Bourgeois, S. T. Treves, and J. Guttag, "Patient-specific seizure onset detection," Epilepsy&Behavior,vol. 5, pp. 483-498, 2004.
[2] J. Dauwels, E. Eskandar, and S. Cash, "Localization of seizure onset area from intracranial non-seizure EEG by exploiting locally enhanced synchrony," in 2009 Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 2009: IEEE, pp. 2180-2183.
[3] D. Chadwick, "Diagnosis of epilepsy," (in eng), Lancet (London, England), vol. 336, pp. 291-5, Aug 4 1990, doi: 10.1016/0140-6736(90)91815-r.
[4] C. Panayiotopoulos, "Clinical aspects of the diagnosis of epileptic seizures and epileptic syndromes," in The epilepsies: Seizures, syndromes and management: Bladon Medical Publishing, 2005.
[5] L. N. Ranganathan, S. A. Chinnadurai, B. Samivel, B. Kesavamurthy, and M. M. Mehndiratta, "Application of mobile phones in epilepsy care," International Journal of Epilepsy,vol. 2, pp. 28-37, 2015.
[6] S. S. Viglione and G. O. Walsh, "Proceedings: Epileptic seizure prediction," (in eng), Electroencephalography and clinical neurophysiology, vol. 39, pp. 435-6, Oct 1975.
[7] Y. Xia and H. Leung, "Nonlinear spatial-temporal prediction based on optimal fusion," IEEE Trans. Neural Networks, vol. 17, pp. 975-988, 2006.
[8] J. Gotman, "Automatic recognition of epileptic seizures in the EEG," (in eng), Electroencephalography and clinical neurophysiology, vol. 54, pp. 530-40, Nov 1982, doi: 10.1016/0013-4694(82)90038-4.
[9] H. Qu and J. Gotman, "A patient-specific algorithm for the detection of seizure onset in long-term EEG monitoring: possible use as a warning device," (in eng), IEEE transactions on bio-medical engineering, vol. 44, pp. 115-22, Feb 1997, doi: 10.1109/10.552241.
[10] A. T. Tzallas, M. G. Tsipouras, D. I. Fotiadis, and Neuroscience, "Automatic seizure detection based on time-frequency analysis and artificial neural networks," vol. 2007, 2007.
[11] W. Weng and K. Khorasani, "An adaptive structure neural networks with application to EEG automatic seizure detection," vol. 9, pp. 1223-1240, 1996.
[12] J. Gotman and L. Wang, "State-dependent spike detection: concepts and preliminary results," Electroencephalography and Clinical Neurophysiology, vol. 79, pp. 11-19, 1991.
[13] L. Guo, D. Rivero, J. Dorado, J. R. Rabunal, and A. Pazos, "Automatic epileptic seizure detection in EEGs based on line length feature and artificial neural networks," vol. 191, pp. 101-109, 2010.
[14] H. Adeli, Z. Zhou, and N. Dadmehr, "Analysis of EEG records in an epileptic patient using wavelet transform," (in eng), Journal of neuroscience methods, vol. 123, pp. 69-87, Feb 15 2003, doi: 10.1016/s0165-0270(02)00340-0.
[15] Z. Zhang, Y. Zhou, Z. Chen, X. Tian, S. Du, and R. Huang, "Approximate entropy and support vector machines for electroencephalogram signal classification," Neural regeneration research, vol. 8, p. 1844, 2013.
[16] 白冬梅, 邱天爽, and 李小兵, "样本熵及在脑电癫痫检测中的应用," 生物医学工程学杂志, vol. 24, pp. 200-205, 2007.
[17] C.-P. Shen et al., "A physiology-based seizure detection system for multichannel EEG," PloS one, vol. 8, p. e65862, 2013.
[18] A. Krizhevsky, I. Sutskever, and G. E. Hinton, "Imagenet classification with deep convolutional neural networks," in Advances in neural information processing systems, 2012, pp. 1097-1105.
[19] S. S. Talathi, "Deep Recurrent Neural Networks for seizure detection and early seizure detection systems," 2017.
[20] R. Hussein, H. Palangi, R. Ward, and Z. J. Wang, "Epileptic seizure detection: A deep learning approach," 2018.
[21] M. Zhou et al., "Epileptic Seizure Detection Based on EEG Signals and CNN," (in English), Original Research vol. 12, 2018-December-10 2018,
[22] Ö. Türk and M. S. Özerdem, "Epilepsy detection by using scalogram based convolutional neural network from EEG signals," vol. 9, p. 115, 2019.
[23] C. Park et al., "Epileptic seizure detection for multi-channel EEG with deep convolutional neural network," in 2018 International Conference on Electronics, Information, and Communication (ICEIC), 2018: IEEE, pp. 1-5.
[24] N. D. Truong et al., "Convolutional neural networks for seizure prediction using intracranial and scalp electroencephalogram," Neural Networks, vol. 105, pp. 104-111, 2018.
[25] P. Boonyakitanont, A. Lek-uthai, and J. Songsiri, "Automatic Epileptic Seizure Onset-Offset Detection Based On CNN in Scalp EEG," in ICASSP 2020-2020 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), 2020: IEEE, pp. 1225-1229.
[26] O. Kaziha and T. Bonny, "A Convolutional Neural Network for Seizure Detection," in 2020 Advances in Science and Engineering Technology International Conferences (ASET): IEEE, pp. 1-5.
[27] 關尚勇, 林吉和, and 陳倩, "腦電圖入門快速判讀," 2014.
[28] P. P. Ngoc, V. D. Hai, N. C. Bach, and P. Van Binh, "EEG signal analysis and artifact removal by wavelet transform," in 5th International Conference on Biomedical Engineering in Vietnam, 2015: Springer, pp. 179-183.
[29] P. Goupillaud, A. Grossmann, and J. Morlet, "Cycle-octave and related transforms in seismic signal analysis," Geoexploration, vol. 23, pp. 85-102, 1984.
[30] I. De Moortel, S. Munday, and A. W. Hood, "Wavelet analysis: the effect of varying basic wavelet parameters," Solar Physics, vol. 222, pp. 203-228, 2004.
[31] C. K. Chui, An introduction to wavelets. Elsevier, 2016.
[32] G. Strang and T. Nguyen, Wavelets and filter banks. SIAM, 1996.
[33] W. S. McCulloch and W. Pitts, "A logical calculus of the ideas immanent in nervous activity," The bulletin of mathematical biophysics, vol. 5, pp. 115-133, 1943/12/01 1943, doi: 10.1007/BF02478259.
[34] A. F. Agarap, "Deep learning using rectified linear units (relu)," 2018.
[35] 斎藤康毅, "Deep Learning：用Python進行深度學習的基礎理論實作," 2017.
[36] F. Seide, G. Li, and D. Yu, "Conversational speech transcription using context-dependent deep neural networks," in Twelfth annual conference of the international speech communication association, 2011.
[37] S. Ruder, "An overview of gradient descent optimization algorithms," 2016.
[38] D. E. Rumelhart, G. E. Hinton, and R. J. Williams, "Learning representations by back-propagating errors," nature, vol. 323, pp. 533-536, 1986.
[39] N. Srivastava, G. Hinton, A. Krizhevsky, I. Sutskever, and R. Salakhutdinov, "Dropout: a simple way to prevent neural networks from overfitting," The journal of machine learning research,vol. 15, pp. 1929-1958, 2014.
[40] V. Dumoulin and F. Visin, "A guide to convolution arithmetic for deep learning," 2016.